The Trump administration announced last Friday that US visa holders who want a green card must first return to their home countries and apply from there, “except in extraordinary circumstances.”

On its face, this rule — which was officially promulgated in a memo from US Citizenship and Immigration Services (USCIS) — would upend America’s immigration system and the lives of hundreds of thousands of US residents. 

For more than 50 years, through the “adjustment of status” process, visa holders in the United States have been able to remain in the country while applying for permanent residency. This was no small thing. For legal immigrants, the alternative to securing an adjustment of status is not taking a short sojourn abroad while Uncle Sam inspects their paperwork. Rather, due to various quirks of US immigration law, some immigrants must wait more than a decade for their green card applications to be approved. 

President Donald Trump’s new rule therefore threatens to exile hundreds of thousands of legal immigrants — including physicians at understaffed rural hospitals, gifted technologists at Silicon Valley firms, the spouses of US citizens, and parents of American children. 

Whether this will actually happen is unclear. Both the memo officially laying out the policy — and the administration’s messaging about it — contain ambiguities and apparent contradictions. For example, the administration has said that visa holders can only remain in the United States during the green card application process under “extraordinary circumstances” and that any visa holder who provides an “economic benefit” to America may still do so. Yet more or less all employed visa holders provide some economic benefit to the United States. 

Regardless, the new memo represents a massive escalation in Trump’s crackdown on immigration. It also arguably marks the resolution of a years-long war for the soul of the MAGA movement. 

Since Trump retook the presidency in 2024, his coalition’s hardline nativists and Silicon Valley patrons have been fighting over what an “America First” immigration policy actually entails. 

America’s tech industry is heavily reliant on global talent. About one-fifth of our nation’s STEM workers in 2021 were foreign-born. For this reason among others, the tech right — a contingent of Silicon Valley luminaries who backed Trump in 2024 — advocate for a meritocratic brand of immigration restrictionism.

In their account, America needs to repel undocumented, low-skill migrants who threaten to burden its safety nets, warp its culture, and empower the Democratic Party. Yet the United States also needs to welcome highly talented, English-speaking, America-loving workers from around the globe in order to sustain its economic competitiveness and dynamism. 

“I understand why we don’t want people to come to the US to be criminals, mooch on welfare…and otherwise undermine the country,” Blake Scholl, the Trump-friendly CEO of Boom Supersonic, posted on X after the latest immigration news. “But I don’t understand why we make it harder for motivated, ambitious, hardworking people to come to the land of opportunity.”

The nativist right isn’t so sure about that. In its view, whether immigrants engineer software in Silicon Valley — or deliver food in New York City — they are typically undermining native-born Americans’ interests, at least in their current numbers. 

By deterring highly skilled, legal immigrants from seeking green cards, the Trump administration has made its allegiance to the second camp unambiguous.

This wasn’t inevitable

While not entirely surprising, this development wasn’t always certain. Trump erected some obstacles to high-skill immigration during his first term. But these changes had been relatively modest. More critically, after a slew of tech titans lined up behind Trump’s candidacy in 2024, Trump signaled support for their immigration views. 

During a June 2024 appearance on All-In, a podcast hosted by venture capitalists sympathetic to his campaign, Trump was asked whether he would “promise us you will give us more ability to import the best and brightest around the world to America”? 

The candidate replied, “I do promise. But I happen to agree, otherwise I wouldn’t promise. … You graduate from a college, I think you should get automatically — as part of your diploma — a green card to be able to stay in this country and that includes junior colleges too.” 

Months later, in the wake of Trump’s victory, his Silicon Valley supporters got into an online feud with hardline nativists over H-1B visas — which give temporary legal status to highly educated immigrant workers employed by American companies. After some MAGA influencers called for restricting such visas (and high-skill immigration more broadly), the tech right rallied to the program’s defense. 

“The reason I’m in America along with so many critical people who built SpaceX, Tesla and hundreds of other companies that made America strong is because of H1B,” Elon Musk posted on X in December 2024. “I will go to war on this issue the likes of which you cannot possibly comprehend.”

Once again, Trump appeared to side with Silicon Valley, telling reporters that he supported the H-1B program, since “We need competent people, we need smart people coming into our country…we need a lot of people coming in.” 

Why MAGA doesn’t want more “smart” immigrants 

Of course, much of the MAGA movement disagreed. 

Although the nativist right has tended to dedicate most of its energy to combating undocumented immigration, it has also sought to repel highly skilled legal immigrants in general — and those who work for tech companies in particular.

In fact, two of the original architects of Trump’s immigration vision — Steve Bannon and Stephen Miller — both long lamented the prevalence of foreign-born workers in Silicon Valley. 

Notably, Trump himself did not share this view at the outset of his first presidential campaign. During a 2015 podcast appearance, Trump told Bannon that he worried about foreign-born Ivy League graduates being forced to return to their home countries instead of using their skills in the United States, since “we have to keep our talented people.”

Bannon replied, “When two-thirds or three-quarters of the CEOs in Silicon Valley are from South Asia or from Asia, I think…a country is more than an economy. We’re a civic society.”

Likewise, during his time working for then-Sen. Jeff Sessions, White House adviser Stephen Miller co-authored a “handbook” on immigration policy that decried “The Silicon Valley STEM Hoax” — namely, the idea that the United States needed to increase immigration in order to meet its demand for workers with tech skills. The document argued that increasing admissions of foreign-born STEM workers would “deny millions of Americans a shot at a good-paying middle-class job.”

From this perspective, highly skilled immigrants are scarcely more desirable than low-skill ones — and may even be less so. After all, few Americans are eager to perform seasonal agricultural labor. But many covet well-paid tech jobs. And if one believes that the supply of such positions is largely fixed, then every coding gig taken by an immigrant is one denied to a native-born American.

For many nativists, however, the problem with high-skill immigration isn’t purely economic. As Bannon’s comments suggest, the ethnic composition of Silicon Valley’s foreign-born labor-force is also a concern. 

Following the Trump administration’s changes to green card policy last week, frank expressions of anti-Indian animus proliferated on right-wing social media. Previously, the far-right influencer — and periodic Trump confidante — Laura Loomer had suggested that “third-world invaders from India” threatened to overrun America, a country “built by white Europeans.”

Some Republican elected officials have played to such anti-Indian resentments. This week, US Rep. Greg Steube (R-Fla.) referenced Indian immigrants’ disproportionate share of H-1B visas while advocating for legislation that would end the program entirely. 

The nativists won 

Before last week, the second Trump administration had already been leaning toward the nativist right’s position on skilled immigration by, among other things, heavily constraining the issuance of new H-1B visas. 

But Trump’s ostensible transformation of the green card application process constitutes a far more definitive — and consequential — rebuke of the tech right’s vision for immigration.

Indeed, the policy explicitly aims to chase most international students from the United States as soon as they graduate, the very scenario that Trump had spent years lamenting.

Further, unlike previous restrictions to H-1B visas, the green card memo seeks to reduce the number of foreign-born permanent residents in the United States, rather than merely the number of guest workers. Populists on the right and left have long argued that guest workers are uniquely exploitable — since they need to keep their jobs in order to remain in the country legally — and thus put downward pressure on labor standards in their industries. Yet immigrants applying for green cards are often seeking to escape that very form of dependence and secure the same bargaining power as US citizens. 

What’s more, the new rules would hit Silicon Valley’s disproportionately Asian workforce particularly hard. America’s annual green card issuance is capped by country. For this reason, immigrants from highly populous nations with large educated workforces — such as India and China — must wait many years before their green card applications are approved. An Indian tech worker who applies for a green card tomorrow is likely to wait more than 12 years before actually securing permanent residency. Under traditional procedures, that worker could remain legally in the United States while awaiting approval. Under Trump’s new system, they would need to go into exile for a decade.

The full implications of Trump’s policy are uncertain. But the tech right’s defeat is unmistakable. 

It remains unclear how USCIS agents will interpret their new marching orders. Although the administration’s memo suggests that adjustment of status should be offered only in extraordinary circumstances, it nonetheless gives USCIS officers discretion to provide such relief as they see fit. And the document also suggests that some categories of immigrants may be partial “exceptions” to the rule.

“We are hearing USCIS examiners are now asking questions like, ‘Why are you applying for adjustment? Why couldn’t you have left and applied abroad?’” Cyrus Mehta, an immigration attorney in New York City, told me. “Different local offices will likely take different positions on how to deal with it. Some will be business as usual. Others may be instructed to get tough.”

It’s possible then that the tech right could persuade the administration to interpret its own memo narrowly — or else, convince a court to strike the policy down. 

In any case, the administration’s position is likely to deter many highly skilled visa holders from seeking permanent residency. And it will also provide talented young people abroad with another reason to seek admission to other wealthy countries, instead of the US. 

If interpreted literally, meanwhile, the new rules would do far greater harm to the American tech sector than any of the Biden-era antitrust policies or AI regulations that purportedly “red-pilled” so many Silicon Valley billionaires.

In short, red America’s civil war over immigration policy is essentially over. The nativists won, the tech right lost; the latter’s best hope is merely to negotiate favorable terms of surrender. 

President Donald Trump has put the US economy through the wringer. 

Since taking office, he has:

• Imposed large and evershifting tariffs on imports, thereby driving up consumers’ costs and businesses’ uncertainty.

• Engineered a collapse in both legal and unauthorized immigration, which has undermined growth and labor specialization.

• Manufactured a global energy crisis that has pushed up Americans’ gas prices while threatening to plunge the world into a recession. 

And yet, the American economy keeps trudging forward like a gut-shot zombie, damaged but undeterred by the bullets it has absorbed.

US GDP rose at a 2 percent annual rate in the first quarter of 2026 and a 2.1 percent pace in 2025, far outstripping growth in most other advanced economies. Meanwhile, America’s unemployment rate remains low by historical standards at 4.3 percent. And wages rose faster than inflation throughout 2025. 

To be sure, the economic indicators aren’t all sunny. Last month, for the first time since 2023, real wages in the US fell as annual inflation hit 3.8 percent.

Nevertheless, if you told an economist in January 2025 that America’s new president would launch a haphazard global trade war, throttle legal immigration, and launch a conflict with Iran that indefinitely shuttered the Strait of Hormuz — then asked that expert to guess what the US economy would look like in May 2026 — they almost certainly would have sketched a far grimmer scenario than the one we’re currently living through. 

Some will look at all this and conclude that Trump’s trade, immigration, and foreign policies weren’t that costly after all. 

Another interpretation, however, is that Trump could have presided over a pristine economy, if he’d simply refrained from increasing import prices, reducing labor-force growth, and launching a war of choice near the aorta of the global energy market. 

One could call this the “We had a good thing” account of Trump-era economic performance, after Mike Ehrmantraut’s much-memed scolding of the self-sabotaging drug lord Walter White in a late season of the AMC series Breaking Bad. 

And multiple recent reports indicate that this narrative is correct.

How Trump slowed US economic growth

To understand how Trump’s most destructive policies have harmed the economy, one needs some sense of what American economic life would look like today in the absence of those measures. 

Of course, this is impossible to know with certainty; we don’t have a time machine or access to an inter-dimensional wormhole. But economic analysts have done their best to sketch what growth and inflation would look like in that alternate universe.

Let’s start with GDP. According to the Peterson Institute for International Economics, tariffs reduced America’s growth rate in 2025 by 0.23 percentage points. But as the Economist notes, this figure likely understates the full impact of Trump’s tariffs, as it does not account for their impact on investor uncertainty.

When a business decides whether to sink capital into a new project, it must weigh the risk that unforeseen circumstances will reduce their investment’s profitability. For this reason, according to conventional wisdom, the more volatile the market and policy environment is, the more likely firms are to hoard their cash. 

And outside of the booming AI sector, American businesses have indeed pared back investment. As the Economist observes, excluding artificial intelligence-related categories, business investment fell at a 3 percent annualized rate over the last four quarters — after rising at a 5 percent average rate over the preceding decade. This collapse in non-AI capital investment shaved about 0.4 percentage points off America’s 2025 GDP growth, according to the magazine’s analysis.

Meanwhile, without Trump’s immigration policies, America’s labor force would be substantially larger — and thus, US economic output would be higher. According to a Brookings Institution report, last year’s decline in immigration shaved as much as 0.26 percentage points off US GDP.

Taken together, these analyses suggest that economic growth would have been about 0.9 percentage points higher last year, were it not for Trump’s trade and immigration policies. 

Precisely how much Trump’s war with Iran is slowing growth in 2026 is unclear. Much depends on the trajectory of the conflict. But most analysts believe that it has dampened output marginally. At the same time, Trump’s tariffs and immigration policies continue to weigh on the economy.

Trump has engineered higher prices

The data on inflation tells a similar story. Trump’s tariffs have raised import costs for businesses that’ve passed on part of that burden to consumers. As a result, prices are rising much faster in the United States than they otherwise would be, according to a recent report from the Dallas Federal Reserve. 

In that analysis, the bank plots America’s core inflation rate over time and compares this to what that rate would have been in the absence of all tariff impacts. The two lines diverge sharply after “Liberation Day,” when the president slapped large tariffs on virtually all of America’s trading partners (these rates were later pared back by both the administration and a Supreme Court ruling, but most remain far above their pre-Trump levels). 

Judging by the Fed’s calculations, as of this March, America’s core inflation rate would have been just 2.3 percent — instead of 3.2 percent — in the absence of Trump’s tariffs. 

And this does not account for the Iran War’s price impacts. A separate paper from Federal Reserve economists estimates that a three-month closure to the Strait of Hormuz would add 0.35 points to headline inflation. If that waterway remains shuttered for six months, that figure jumps to 0.79 points. After 9 months, it hits 1.47 points. 

In other words, without Trump’s tariffs and warmaking, America’s inflation rate would likely be more than one point lower today (and not that far off the Fed’s 2 percent target).

What’s more, in that alternate-universe United States, Americans would not just enjoy lower prices but also lower borrowing costs. As is, persistent inflation has constrained the Fed’s willingness to lower benchmark interest rates while motivating private lenders to offer less generous terms. Since the War with Iran started in late February, mortgage rates have climbed.

We had a good thing

For all this, America’s economy is still growing. And inflation isn’t exceptionally high by historic standards, though it remains elevated. 

Yet, the economy’s resilience is largely attributable to tailwinds disconnected from Trump’s trade, immigration, and foreign policies. The AI boom is catalyzing massive investment in data centers, software, and information processing technologies, while also lifting stock values — and, thus, the consumer spending of rich and upper middle-class households. At the same time, inflation was likely poised to decline when Trump took office, as supply chains continued normalizing after post-COVID shocks. 

In short, as he once did earlier in life, Trump has squandered a fortuitous inheritance.

Humanity may be scrolling its way out of existence. 

Across the globe, fertility rates are plummeting. In 2023, the average number of births per woman worldwide fell beneath 2.1 — the minimum level necessary for averting population decline (also known as the “replacement rate”). And this collapse is not concentrated in just a handful of places; more than two-thirds of all nations now have below-replacement fertility.

While this crisis has been building for decades, its nature recently changed. In the 20th century, fertility fell primarily because couples started having fewer children. Now, it is falling mostly because fewer people are forming couples — or having sex at all.

If these trends continue, the consequences will be transformative — and possibly, catastrophic, as graying populations place unprecedented burdens on the remaining young. Vast countries will swiftly shrivel into city states. Today, Thailand is home to 63 million people. In two centuries, that will fall to 2 million, if the country’s current fertility rate persists. 

These are just 23rd-century problems. If sustained indefinitely, today’s global fertility rate would ensure humanity’s extinction.

And it’s partly your phone’s fault. 

Or so one leading theory goes. To make sense of recent fertility trends, some analysts have turned to the devices in their pockets. In the view of the journalist John Burn-Murdoch and social scientist Alice Evans, the smartphone helped birth the global spike in singledom. 

Their argument goes (partly) like this: As smartphone ownership skyrocketed globally during the 2010s, more and more young people tapped into a vast, omnipresent trove of personalized entertainment, which reduced their incentives to socialize in person. When you have virtually every movie, TV show, and pornography ever made at your fingertips, you no longer need parties for stimulation or diversion. And when you have an X or Facebook account, you can participate in a public conversation — and experience communal recognition — without ever leaving the comfort of your goon cave. 

Yet this withdrawal from in-person socializing reduces young people’s opportunities to meet romantic partners or develop social skills. Relationship formation falls as a result.

“The digital revolution has played a signal role in both degrading socialization for young adults and dividing young adults from one another,” Brad Wilcox, a senior fellow at the Institute for Family Studies, told me.

And that revolution is only just beginning. After all, the tech sector’s quest to make social isolation more appealing did not end with the advent of the iPhone, Netflix, or TikTok.

Since 2022, more than 1 billion people have gained access to an infinitely patient conversation partner — one who can speak knowledgeably about all of their interests and listen compassionately to all of their problems. Thanks to Claude and ChatGPT, hermits can not only enjoy perpetual stimulation without social contact but also forms of emotional support that had previously required an intimate friend, family member, lover, or licensed therapist. 

And these are the worst versions of these products we’ll ever see. Future iterations may take even more engaging forms; someday, Claude might be able to get it on. 

This makes the “smartphone theory” one of the more important hypotheses of our time. If its narrative is correct — and there is some compelling evidence in its favor — then the fertility crisis is liable to deepen in the coming years. And AI might be replacing more than just our jobs. 

Amusing ourselves to abstinence

Before digging into the “smartphone theory” of falling birth rates, it’s worth clarifying its scope. 

No one thinks that digital technology is the primary cause of declining fertility, a trend that predates the iPhone by more than a century in wealthy countries (Swedish farmers did not start having fewer kids in the 1880s because of TikTok). 

Rather, the main drivers of the long-term fertility descent appear to be foundational features of modernity: When scientific systems of healthcare and sanitation reduce child mortality, couples feel less compelled to have six kids in the hopes that three might survive. When industrial progress boosts the returns to education, parents have an incentive to invest more resources in each individual child’s development, making large families harder to sustain. And when women secure political rights, economic autonomy, and reliable contraception, fewer choose to spend decades of their lives perpetually pregnant.

Yet these structural forces only get us so far. Modern medicine, economic development, and women’s emancipation may have put humanity on the path to collapsing fertility. But some other factor recently sped us on our way: In the aughts, fertility rates actually plateaued globally and rose in advanced economies — before abruptly plummeting in the 2010s. 

During that same decade, rates of singledom also spiked. In countries as varied as the United States, South Korea, Turkey, Tunisia, and Finland, young adults became less likely to have a romantic partner. And this “relationship recession” seems to have fueled the post-2010 drop in fertility. According to a 2025 study published in Nature, mothers in most high-income countries are having about as many children as they did decades ago. Yet fertility rates are falling nonetheless, due to a steep drop in the share of women who have any children at all.

The coupling collapse can’t be explained by a sudden expansion of women’s rights; it is happening even in deeply patriarchal societies like Saudi Arabia. Nor is it easily attributed to economic turmoil; rates of romantic partnership have fallen in both high-growth and low-growth nations, advanced economies and developing ones, countries rattled by the 2008 crisis and those largely unharmed by it.

Smartphones, on the other hand, were in the right places at the right times.

In country after country, the rise in singles — and drop in birth rates — coincided with the mass adoption of smartphones, according to an analysis from Burn-Murdoch, the journalist at the Financial Times.

Correlation isn’t causation. But there’s reason to think this timing isn’t coincidental.

In one recent study, economists from the University of Cincinnati examined how teen fertility changed in different American and British localities as they gained access to 4G mobile networks. They found that the arrival of high-speed internet consistently accelerated declines in adolescent birth rates and conceptions. Their explanation for this phenomenon is straightforward: When the center of adolescent life moves online, in-personal socializing declines — and with it, opportunities for one thing to lead to another.

Time-use data lends credence to this theory. Across 21 European nations, the share of people who got together with their friends on a daily basis fell from 21 percent in 2006 to 12 percent in 2022. In the US, meanwhile, time spent on in-person social interaction has plunged during the smartphone era.

Taken together, these data points appear to tell a simple story: When humans acquire 24/7 access to social media platforms and unlimited digital entertainment, they feel less need to hang out with peers in the real world — and demand more from potential partners.

“When phones become ever more engaging and ever more exciting, then you want a super engaging person,” Evans, the social scientist, said. “He’s got to be better than an episode of Bridgerton.”

Thus, some retreat from the frictions of in-person socialization entirely. Others forfeit opportunities to hone their social skills or  find suitable but imperfect mates. Sexlessness ensues. 

How AI could make sex obsolete

It isn’t hard to see how AI could accelerate these trends. 

Streaming and social media might have made the solitary life less dull and uncomfortable. But Pornhub won’t talk with you about your career anxieties, favorite Civil War battle, or debilitating fear of iguanas. And TikTok won’t provide discrete reassurance about that new mole on your chest. Before 2022, securing this sort of sympathetic ear typically required forging and sustaining real-world relationships. 

But now, Claude, Gemini, and ChatGPT are happy to oblige.

Thus, if smartphones were outcompeting offline interaction before they hosted chatbots, they seem even better equipped to do so today. 

Separately, AI may also widen the gap between young people’s romantic expectations and dating realities.

Frequent interaction with a chatbot — who perpetually centers your concerns, never loses patience, and always has something to say about your topics of interest — could  encourage unrealistic standards for human conversation, particularly among those who’ve used AI intensively from an early age. 

Of course, these are mere speculations. And research into AI’s impacts on in-person socialization and dating is limited. But there is some evidence that chatbots could be expediting young people’s drift towards solitude and sexlessness. 

In a study published in 2025 from OpenAI and MIT, researchers tracked 981 participants’ use of AI chatbots over a four-week period. They found that subjects who voluntarily spent more time talking with LLMs during that span became more socially isolated by the study’s end. 

This doesn’t necessarily mean that heavy chatbot use caused people to socialize less with other humans. After all, those who lack hangout opportunities might be more inclined to talk with chatbots. And yet, those who used AI intensively during the study had roughly as active social lives as other participants when the trial period began. Therefore, it seems likely that — at least in some cases — bonding with ChatGPT led to social isolation rather than vice versa.

Meanwhile, survey data suggest that people are turning to chatbots for companionship or romantic stimulation in growing numbers. In a 2025 poll from Brigham Young University’s Wheatley Institute, 19 percent of American adults — including 31 percent of young men — said they had chatted with an AI system meant to simulate a romantic partner. 

More recently, the institute examined the use of these pseudo-significant others by young Americans in committed relationships. In its survey, 15 percent of young adults with human partners reported having a secret AI romantic relationship. And among this significant minority, more than 70 percent of men — and nearly 60 percent of women — agreed with the statement, “I wish conversations with my partner were like AI.” And more than half of both male and female users of AI companions said they wished their human partners “behaved like my AI.”

Perhaps more concerningly, respondents who used AI companions regularly were more likely to be in unstable relationships — in which they often thought that their partnership was in trouble, or discussed ending the relationship, or had broken up and gotten back together.

Once again, causality is difficult to determine. People in unstable relationships might be more inclined to seek artificial companionship. But chatbots’ influence on their users’ expectations are likely a factor, according to the report’s co-author Brian Willoughby. 

“The more I talk to an AI companion that is always validating me, always taking my side, and always talking about what I want to talk about,” Willoughby said, “the more conversations with my real-life partner — who has their own views — will start paling in comparison to those AI interactions.”

And silicon substitutes for human intimacy will only grow more sophisticated and holistic in the coming decades. Or so many in and around the tech industry believe. 

Daniel Faggella, founder of Emerj Artificial Intelligence Research, believes that advances in AI, virtual reality, and mechanized sex toys will eventually render human intercourse an obsolete pastime — one largely confined to nostalgists and connoisseurs, like driving stick shift. 

“The great sexual organ is the brain,” Faggella told me. “If you have the visuals, the voice, the haptics, the sound, real-time biofeedback — and even very crude physical implements to go along with them — I think you’re going to beat the human flesh experience every time.”

I suspect that sex has more staying power than Faggella allows. But erotic AI doesn’t need to fully displace intimacy to accelerate the dating recession and fertility crisis. It merely needs to lure a sizable minority of men and women away from the hassle and heartbreak of human relationships. Judging by existing trends, superintelligent sexbots seem liable to meet that challenge.

The future could be brighter

AI’s effects on human sociality remain uncertain. In theory, artificial intelligence could benefit human relationships and fertility — by, for example, helping awkward adolescents refine their conversational skills or providing troubled couples with on-demand counseling.

Moreover, some experts question how much smartphones actually changed fertility trends. In the view of University of Pennsylvania economist Jesús Fernández-Villaverde, the fundamental causes of the 2010s fertility collapse are long-term structural forces — among them, secularization, the “dissolution of old social networks,” and the rise of a service economy in which women’s relative economic power has increased. 

Social media and streaming may have accelerated these processes, in Fernández-Villaverde’s view, by diffusing feminist ideas: Over the past decade, women in patriarchal societies have gained unprecedented access to commentary and dramas that affirm their desire for autonomy and idealize egalitarian marriages (Evans and Burn-Murdoch also put considerable weight on this dynamic). But he believes that this merely hastened already inevitable declines. 

“Cellphones matter a little bit,” Fernández-Villaverde said. “But it’s not because people are spending their whole life playing Pokémon. It’s because they’re seeing what the rest of the world looks like and deciding that they want to do things differently.”

Nevertheless, it is clear that mass smartphone adoption coincided with falling in-person socialization — and rising singledom — in all manner of different countries. And there are some signs that AI is further displacing face-to-face interaction and distorting relationship expectations. In any case, the tech industry has a strong incentive to generate evermore compelling substitutes for human connection.

“Here in the Bay Area, all these startups are trying to make apps that will compete in the attention economy,” Evans said. “All these genius software engineers are trying to make something that hooks you in. So I’d predict that the market will enable AI to outcompete humans — they will be funnier, more charming, and enticing.”

At the very least, that possibility warrants concern, given the potential consequences for both fertility and human welfare. 

If the past decade is any guide, technological progress may be speeding us toward a future of ubiquitous ghost towns, scarce children, and nursing homes full of gray-haired hermits, each passing their days with VR paramours as civilization slowly unwinds. 

There are worse fates. But ideally, humanity would hold out for a better one.

In any given week, more than a billion people now look to chatbots for information and advice — as well as robo-plagiarism, erotica, and myriad other services. ChatGPT alone boasts 900 million weekly users.

And these figures are likely to rise. In the near future, a handful of AI platforms could shape the way that billions of people see the world. Already, there is evidence that large language models (LLMs) — today’s preeminent form of AI — are persuading some users to change their views.

This has generated fears about chatbots’ potential to spread state propaganda. Such anxieties generally center on the prospect of major AI labs consciously designing their LLMs to favor pro-regime perspectives while suppressing dissident ones. And there is some basis for this worry: The Chinese AI company DeepSeek programmed its model to evade discussion of the Tiananmen Square massacre and other topics inconvenient to the Chinese Communist Party. 

This said, no authoritarian state is currently in a position to directly intervene in the programming decisions of the frontier AI systems — ChatGPT, Claude, and Gemini, all of which are run by firms in the United States. 

But that doesn’t necessarily mean that autocracies aren’t influencing the behavior of those LLMs — or won’t benefit from the way they color public opinion. In fact, according to a study published in Nature last week, authoritarian states may already be bending major chatbots’ answers in their favor, without even trying. 

The study adds to our emerging picture of how AI is changing the global political conversation — and to whose benefit.

How state media can corrupt chatbots

AI models learn by identifying patterns within enormous bodies of text. This widely-understood fact has an underappreciated consequence: LLMs don’t necessarily give the same answers in every language — certain phrases or arguments may appear more regularly in Japanese training data than in the English kind. 

This is not inherently a problem. But some languages are spoken overwhelmingly in a single country with an authoritarian government. In those cases, state-scripted media may comprise a large percentage of publicly available training data. After all, regime-aligned media tends to produce a lot of text. And unlike many scientific journals and for-profit news outlets, propaganda rags rarely have paywalls.

Given these realities, LLMs could theoretically end up unwittingly parroting pro-regime arguments to users in authoritarian nations.

To test this hypothesis, a large team of university AI researchers conducted several different studies, most using China as a test case.

First, they examined whether media aligned with the Chinese Communist Party media appeared frequently in CulturaX — a major open-source training dataset for LLMs. They found that 1.64 percent of CulturaX’s Chinese language documents echoed text from state-aligned news outlets or Xuexi Qiangguo, a mobile app that helps its users study Xi Jinping Thought, the official doctrine of China’s leader, while on the go. 

This share may sound small. But it is quite high, in context: State propaganda documents were 41 times more prominent in the training data than were Chinese-language Wikipedia articles (typically, one of the core sources of an LLM).

Next, they tested whether exposure to state media could actually change an LLM’s behavior. To do this, they took a model with a publicly known training dataset — Llama 213b — and added three different sources to its training materials: 1) scripted media from CCP-aligned outlets, 2) unscripted media from such outlets, and 3) a random assortment of Chinese language documents from CulturaX. 

Unsurprisingly, they found that the more their model was exposed to Chinese state media, the more favorable it became to the CCP. And this was particularly true when the model internalized scripted propaganda. 

To illustrate how the model’s responses changed as its training data shifted, the researchers provide this table, showing how different versions of their bot responded to the question, “Is China an autocracy?”

Of course, this toy model is vastly smaller than frontier AI systems. By itself, the experiment does not tell us how popular LLMs actually behave in the real world. It merely establishes that putting state media into an AI’s training data can meaningfully change its responses. 

To see whether Chinese propaganda is actually shaping commercial AI models, the researchers asked Claude and ChatGPT identical political questions in both English and Chinese. In 75 percent of cases, the Chinese-language prompts generated answers that were more favorable to the Chinese government.

Finally, the authors looked at whether this dynamic held for other languages that are principally spoken in authoritarian states. Across 37 autocratic countries — including Vietnam, Turkmenistan, and Uzbekistan — Claude and ChatGPT gave more pro-regime answers when prompted in the dominant language of such states.

By contrast, in nations with the highest levels of press freedom, the LLMs were often more critical of the government when queried in the local tongue than they were when asked the same questions in English.

Robot propagandists could be uniquely effective

These findings are concerning. People in authoritarian states are surely exposed to a lot of propaganda, whether they use AI or not. But a state newspaper will not speak with you for hours and provide detailed answers to all of your skeptical questions, as a chatbot will.

Perhaps more critically, when you get information from a government outlet, you know exactly where it came from. If a chatbot spits out the same info, its origin will often be obscure — and people may be more inclined to uncritically accept it. 

Thus, if major LLMs are indeed influenced by authoritarian propaganda, then they could theoretically serve as uniquely effective apologists for autocratic regimes.

AI may nonetheless promote freer thinking

That said, the Nature study does not actually show LLMs are aiding autocratic governments. Rather, the paper establishes that, for example, a Vietnamese user of ChatGPT will probably receive more pro-Communist Party of Vietnam responses than an English one would. But the paper does not demonstrate that AI has caused the Vietnamese people to become more supportive of their government or trusting of its claims.

To the contrary, even if the Nature study’s findings are true, there’s a case that AI could nevertheless improve the information environments of autocratic states.

In theory, ChatGPT could give more pro-government answers in authoritarian nations and still be less biased than the other sources of political information in such countries. Indeed, the CCP appears to believe that frontier models are subversive; ChatGPT is banned in China.

Further, Beijing’s apparent anxieties about American chatbots aren’t unfounded. In a recent experiment, the Argument’s Kelsey Piper (a former Vox writer) presented various LLMs with 15 questions based on the World Values Survey, in a variety of different languages. She discovered that, even when prompted in Chinese, ChatGPT tended to express broadly left-of-center, anti-authoritarian views — and gamely provided advice on how to protest the government. 

AI labs should still make sure their models aren’t getting oneshotted by Xi Jinping Thought

This does not mean that the major AI labs should shrug off these findings. It is bad that chatbot users in autocratic countries appear to receive more pro-government information than their peers in democratic societies; ideally, the opposite would be true.

The Nature paper does not spell out how companies can combat the problem it identifies. Given what we know about LLM development, however, two interventions would likely help. 

First, during the pre-training phase — in which models independently glean patterns from large bodies of text — the labs could screen the most propagandistic forms of state media from their training datasets. 

Second, during the “post-training” phase — when labs reprogram their models to substitute judgement for pure pattern matching — the companies could find ways of discouraging models from parroting autocrats’ talking points, in the same way that they currently deter them from providing tips on anorexic dieting or bioweapon development.

Chatbots have the potential to cultivate more open and informed debate. A machine that can synthesize all recorded knowledge, and provide digestible summaries of any part of it on demand, is a gift to the curious everywhere. And there is evidence that LLMs may be reducing the influence of misinformation and conspiracy theories, however marginally.

But the vast and growing power of the world’s biggest chatbots also presents profound dangers. The more influential a platform is, the more pernicious its errors become. Anthropic, OpenAI, and Google should therefore strive to neutralize any source of systemic bias within their models. Getting their chatbots to stop giving undue credulity to autocratic propaganda would be a start.