Last Saturday, I paid a visit to a little slice of the resistance in its current, attenuated form. Since Donald Trump’s second inauguration, there have been many street protests, but they have been small and diffuse: a few hundred people angered by the defunding of USAID, or a couple thousand in support of national parks. The most organized effort so far, the 50501 movement (for “50 states. 50 protests. 1 movement.”), is a coalition of activists whose name telegraphs breadth much more than depth.
At an intersection outside a Tesla showroom in the Gowanus neighborhood of Brooklyn, about 200 people were gathered on the sidewalk. They banged tambourines and clanged cymbals, trying to get passing trucks and cars to honk against Elon Musk, Trump’s most important adviser and a purveyor of electric cars. Of all the responses to the new administration, the anti-Tesla protests have left a bruise—as of this writing, the company’s stock has declined by about half since December. There were signs demanding Musk’s deportation, and one that asked drivers to Honk If You Think Elon Is a Dork. Someone in a gorilla suit held up a placard that read He Kills Monkeys, Too.
For all the energy on the street—and it was energetic—there was a preponderance of gray hair, and not many young people or strollers in evidence. Those who showed up seemed to have a highly developed muscle memory for activism, going back perhaps to the anti–Iraq War demonstrations in 2003, or even Vietnam. This appeared to be a protest by and for a committed core. Maybe the most revealing poster was one that declared The Protests Will Get Bigger Until the Constitution is Respected. This was a threat, of course, but also an acknowledgment that there was plenty of room to grow.
Nevertheless, for those in the exhausted anti-Trump coalition, these bursts of opposition are giving hope. Today, a coalition of liberal groups under the banner “Hands Off!” is planning hundreds of such actions around the country. This is the kind of activity that has led members of Harvard’s Crowd Counting Consortium, which tracks acts of civil dissent, to conclude recently that the resistance is “alive and well,” with protests “far more numerous and frequent than skeptics might suggest.” They have what at first glance seems like an unexpected finding: In February 2025, twice as many protests took place as in February 2017, during the tumultuous beginning of Trump’s first term: 2,085 versus 937. The major caveat is that what they count as a protest event could be a couple of people handing out flyers on campus. Altogether, the number of actual protesters is far, far below what it was eight years ago.
The best estimate for this past February was somewhere from 125,000 to 184,000 participants, according to Jeremy Pressman, one of the consortium’s co-directors and a professor at the University of Connecticut. This would put the average protest size anywhere from 60 to 88 people. In March, Pressman said, those numbers increased significantly, but the per-protest average stayed roughly the same.
What this suggests, at best, is a different model of protest movement: highly decentralized, moving at a snail’s pace, more a slog than a resistance. “Something is happening,” the journalist Ali Velshi wrote on MSNBC last week, “a different kind of movement building right now, one that has had steady and sustained momentum.” In The Bulwark, Jonathan V. Last emphasized the strategic advantage of a movement that makes its way from the hinterlands toward Washington.
It is different, but is it better? Is a single protest of 100,000 people equal to 1,000 actions with 100 people at each? I posed this zen koan of a question to Erica Chenoweth, a Harvard professor and Consortium co-director who coined the idea that if any protest movement drew 3.5 percent of a country’s population, it could achieve its goals. (That would equal nearly 12 million people in the United States today.) Does it matter how you get to this figure, all at once or bit by bit? “We don’t really know,” Chenoweth told me, “and conceivably either path produces momentum.”
So far into Trump 2.0, though, the path of decentralized slowness has had a paradoxical effect: It’s giving activists lots to do but is leaving a much larger population of dissenters without an expressive outlet. What makes a disaffected Gen Zer or a busy Millennial parent drop what they are doing and head into the streets is very different from what motivates hard-core protesters to pick up their cymbals. That much larger group needs to feel both the safety and the collective impact that comes with a mass march. “Power springs up between men when they act together and vanishes the moment they disperse,” the philosopher Hannah Arendt wrote in The Human Condition. Most people join protests to express that power, not to emphasize their marginality. And it is precisely this type of action that, in the face of one barrier after another, feels more difficult than it ever did.
Trump’s first term was punctuated by a series of monster gatherings: the Women’s March that greeted his inauguration (estimated at as many as 4.6 million people all over the country), the March for Our Lives following the 2018 Parkland shooting (1.2 million), and—dwarfing all previous American protest movements—the Black Lives Matter demonstrations after the killing of George Floyd (anywhere from 15 million to 26 million people, according to polls taken at the time). The Women’s March was highly organized and concentrated, while the protests in June 2020 were largely spontaneous and spread out. But what made all of these significant was the measure by which protest has long been judged: the overwhelming numbers of people who took part.
On its face, the slackening of interest in demonstrating against Trump in his second term—despite the preponderance of issues that trigger his opposition every day—can be understood simply as a change in the political atmosphere. Trump’s 2017 win was a shock for liberals expecting to inaugurate the first female president. His failure to win the popular vote made his election feel contestable, even illegitimate. And powerful people, even a number of Republicans, were ready to stand up to Trump’s overreach. All of this created a sense that there was a door that protest could push open. None of these factors operates in 2025: Trump is not a surprise, he won the popular vote, and both his party and the business elite are fully on his side. Pressman wondered if this has created “a different kind of shock,” one that is more destabilizing than motivating: “Maybe people were just kind of pushed on their back feet. And it takes a while, or ever, to be involved in it.”
After the University of Glasgow lecturer Michael T. Heaney surveyed participants from both the Women’s March and January’s much shrunken People’s March, he emerged with a clear contrast. There was a 12 percent drop in enthusiasm about politics, he told me; “hope” as a motivator for protesting fell by 10 percent, and “pride” by 9 percent. What rose from 2017 to 2025? “Frustration,” by 3 percent, and “anger,” by 8 percent.
Alongside demoralization runs justifiable caution; the act of protest itself has become more dangerous. Not only has the president said that he would have little compunction about using the military to deal with “the enemy from within,” but the tactics of surveillance—including facial recognition, geolocation tracking, and AI-enhanced identification—have gotten more pervasive and sophisticated. The police and the FBI have long used and abused tools for monitoring protest—this sordid history goes back to COINTELPRO in the 1960s. But the average protester at a peaceful 20th-century gathering could at least assume that they would melt into a sea of indistinguishable people. Technology has made that impossible. There is no safety in numbers. You cannot disappear.
Simply being present at a protest makes you vulnerable, as Chris Gilliard, a Just Tech fellow at the Social Science Research Council, told me: “All the devices that people carry and wear are constantly extruding data, from the car they might have used to get there to the cameras in subway stations to their watches and their phones and everything else you can imagine.” Among the safety instructions provided by an organizer of today’s nonviolent demonstrations was the bolded sentence “Do not assume you are safe.” Also: “MASK. UP … disable location, biometrics and data on your phone, at the very least. Have your emergency contacts written on your body.” This is not a welcoming message for an infrequent protester or a citizen looking to voice her concern for the first time. No wonder that a smaller, local protest is more of a draw at the moment. “It’s easier to control; it’s just easier to call it off suddenly if you need to; it’s easier to get away,” Chenoweth told me. “There’s lots of reasons why smaller, more nimble groups would feel more safe.”
There are also an increasing number of laws targeting protest. The volume of federal and state legislation has spiked in the past few years, and particularly since January; 41 anti-protest bills have been introduced in 21 states and in Congress in 2025. This crop includes bills by Congress against protesters “deliberately delaying traffic,” disrupting in any way the construction of a pipeline, or wearing a mask that “oppresses” another person. Among the Trump administration’s demands on Columbia University in return for restoring $400 million in funding was a ban on masks, which many peaceful protesters wear not to menace others but to avoid being identified by facial recognition.
Most of the movements we now consider heroic were originally perceived as lawless nuisances, but Trump has reframed and stigmatized protest in novel ways, conflating nonviolent gatherings with destructive mobs. The reactions to the killing of George Floyd, which did include acts of vandalism and looting, were overwhelmingly peaceful, according to a study by the group Armed Conflict Location & Event Data. But the Black Lives Matter protests of that summer are now remembered by Trump and his supporters as nothing more than a series of riots. The pro-Palestinian activism of the past year and a half has been characterized at times by intimidation, which may have undermined the reputation of protest. But Trump has gone further by painting all legitimate expressions of distress over mass death in Gaza as “pro-Hamas” and moving to deport noncitizens who had anything to do with it. Mahmoud Khalil, the former Columbia graduate student whose deportation is now pending, became an obvious target for the government precisely because he openly protested, unmasked.
As recently as two years ago, the act of picking up a sign and walking out your door to join a march could feel affirming, even joyous. Now it comes with the fear of surveillance and recrimination, the possibility of felony arrest, and, as a result, a deep sense of resignation.
Why does size matter when it comes to protest? Chenoweth’s magic number notwithstanding, the reasons are less tangible than a simple equation. Large demonstrations are a performance of plurality, with individuals transcending their own individuality to express a larger will. This can happen only when the public around you is more than just the person in the gorilla suit who usually shows up to these things. The experience can be bolstering. It can reassure you not just that you aren’t alone, but that you are a citizen among citizens. The sociologist Émile Durkheim had a wonderful phrase for this: collective effervescence.
And for those watching from on high—senators, judges, CEOs—mass protests are physical manifestations of public opinion. They might make a singular person in the throng invisible, but they make their opinion super-visible. Whatever one feels in retrospect about the 2020 BLM protests, every sector of society had to contend with their enormity—it was a moment when even the CEO of JP Morgan Chase, Jamie Dimon, seemed to feel compelled to take a knee. Trump’s crusade to root out DEI is, in many ways, a backlash to the changes these protests spawned. Outside the U.S., consider a recent textbook example of mass mobilization leading to a successful outcome: the 2023 marches in Israel in response to Benjamin Netanyahu’s judicial reforms. These protests effectively shut down the country and forced Netanyahu to relent.
The most optimistic story activists tell about what is happening now is one of incubation: You can’t get to big unless you go through small. Groups are forming coalitions, developing organizational structures, testing the waters locally. This is crucial work for any movement that wants to be long-lasting. One of the major critiques of the protests that blazed so brightly in Trump’s first term is that they largely flamed out very quickly. So this time could be different. Hunter Dunn, a student at Pepperdine University who is also a national press liaison for the 50501 movement, told me that he is a rarity at these protests because he isn’t a Gen Xer or a Boomer. But as a young person fighting what he described as the hopelessness of his fellow young people, he too has visions of the movement culminating in a classic 1963-style march on Washington. “I don’t think we can just get millions of people there tomorrow if we announce that,” he said. “We have to build towards it over weeks, months, even maybe over a year, because it takes that long to build a movement that is strong enough to go to Washington, go there peacefully, but go there unafraid of the government.”
Yet there is also a pessimistic version, one the activists are not telling: that this is a slog without end. Not a slow gathering with a grand finale, but a defanged alternative to mass mobilization; not the prelude to an eruption, but a reliably timed release valve.
There is no reason yet to think that this will be the story of protest in Trump’s second term. Last week, Senator Chris Murphy said that Democratic Party resistance might require “mass-scale mobilization.” This weekend’s protests may well be an early inflection point in that direction—or they may just continue the scattered pattern seen so far. A few activists also pointed me to the large rallies that have greeted Bernie Sanders and Alexandria Ocasio-Cortez on their ongoing “Fighting Oligarchy” tour. What was remarked upon was the size of the crowds—more than 30,000 in Denver—the kind of “surprising number,” as Chenoweth put it, that matters for creating momentum.
On a Reddit post drumming up support for the country-wide April 5 protests—more than 1,000 are planned—one user had their own unequivocal answer to the question of how best to build momentum. “We ask you to make it to the largest planned protest you are able to,” protectresist wrote. “200 people each at 100 protests will not make the news. 20,000 people at one protest will.”
Donald Trump had a plan. It was not a good plan, or even a plausible one. But it was, at least, a coherent plan: By imposing large trade barriers on the entire world, he would create an incentive for American business to manufacture and grow all the goods the country previously imported.
Whatever chance this plan had to succeed is already over.
The key to making it work was to convince businesses that the new arrangement is durable. Nobody is going to invest in building new factories in the United States to create goods that until last week could be imported more cheaply unless they’re certain that the tariffs making the domestic version more competitive will stay in place. (They’re probably not going to do it anyway, in part because they don’t know who will be president in four years, but the point is that confidence in durable tariffs is a necessary condition.)
Trump’s aides grasped this dynamic. “This is the great onshoring, the great reshoring of American jobs and wealth,” Stephen Miller, Trump’s deputy chief of staff, declared on “Liberation Day.” The White House accordingly circulated talking points instructing its surrogates not to call the tariffs a leverage play to make deals, but to instead describe them as a permanent new feature of the global economy.
But not everybody got the idea. Eric Trump tweeted, “I wouldn’t want to be the last country that tries to negotiate a trade deal with @realDonaldTrump. The first to negotiate will win – the last will absolutely lose.”
Eric’s father apparently didn’t get the memo either. Asked by reporters whether he planned to negotiate the tariff rates, the president said, “The tariffs give us great power to negotiate. They always have.”
Someone seems to have then told Trump that this stance would paralyze business investment, because he reversed course immediately, writing on Truth Social, “TO THE MANY INVESTORS COMING INTO THE UNITED STATES AND INVESTING MASSIVE AMOUNTS OF MONEY, MY POLICIES WILL NEVER CHANGE.”
However, there is a principle at work here called “No backsies.” Once you’ve said you might negotiate the tariffs, nobody is going to believe you when you change your mind and say you’ll never negotiate.
Indeed, precisely two hours and 17 minutes after insisting that his policies would never change, Trump returned to Truth Social to announce excitedly that the policies were going to change: “Just had a very productive call with To Lam, General Secretary of the Communist Party of Vietnam, who told me that Vietnam wants to cut their Tariffs down to ZERO if they are able to make an agreement with the U.S. I thanked him on behalf of our Country, and said I look forward to a meeting in the near future.”
The possibility remains that Trump will revert to insisting that the tariffs are permanent and irrevocable. The day is still young.
To be sure, signaling openness to negotiation on tariffs is also a plan. But it’s a very different plan than attracting massive investment into domestic production. The idea behind this other plan is a game of chicken: We think the balance of trade protection is unfavorable to the United States, and could be made more favorable by leveling the playing field. Threatening a global trade war imposes pain on other countries, making them willing to reduce their tariffs on American goods, leading to freer global trade.
This strategy relies not on convincing businesses that Trump is completely serious, but instead on making other world leaders believe that Trump is willing to endure fantastic amounts of economic pain in order to gain bargaining leverage. “Sometimes the best strategy in a negotiation is convincing the other side that you are crazy,” the investor and pro-Trump social-media influencer Bill Ackman rationalized on X.
Could it work? Like the original plan, this is not a good one by any means. Attempting to negotiate new trade deals with almost every country and territory in the world, some of which are uninhabited, poses a formidable diplomatic challenge. (Do penguins have a trade representative?) There is very little in Trump’s record to suggest that he’s going to pull it off. The fact that he is freezing domestic investment decisions in the meantime and risking stagflation or a recession is going to undermine his leverage rather than increase it.
We’re the ones who are waging a trade war against the entire planet. Attempting to intimidate all other countries at the same time is a bit like a school bully walking into the cafeteria and announcing that everybody has to start handing over their lunch money on an ongoing basis. The strong-arm method is best suited for one-on-one negotiations, rather than giving everyone an incentive to band together in self-protection.
In any case, the madman approach to achieving freer global trade by tanking the American economy, whatever odds it may stand of success, is simply not the same thing as creating a permanent new system of protected domestic industry. That vision had danced in Trump’s head since the 1980s. He decided that now was his chance to finally make it happen. It lasted less than a day.
Deep down, Sam Altman and François Chollet share the same dream. They want to build AI models that achieve “artificial general intelligence,” or AGI—matching or exceeding the capabilities of the human mind. The difference between these two men is that Altman has suggested that his company, OpenAI, has practically built the technology already. Chollet, a French computer scientist and one of the industry’s sharpest skeptics, has said that notion is “absolutely clown shoes.”
When I spoke with him earlier this year, Chollet told me that AI companies have long been “intellectually lazy” in suggesting that their machines are on the path to a kind of supreme knowledge. At this point, those claims are based largely on the programs’ ability to pass specific tests (such as the LSAT, Advanced Placement Biology, and even an introductory sommelier exam). Chatbots may be impressive. But in Chollet’s reckoning, they’re not genuinely intelligent.
Chollet, like Altman and other tech barons, envisions AI models that can solve any problem imaginable: disease, climate change, poverty, interstellar travel. A bot needn’t be remotely “intelligent” to do your job. But for the technology to fulfill even a fraction of the industry’s aspirations—to become a researcher “akin to Einstein,” as Chollet put it to me—AI models must move beyond imitating basic tasks, or even assembling complex research reports, and display some ingenuity.
Chollet isn’t just a critic, nor is he an uncompromising one. He has substantial experience with AI development and created a now-prominent test to gauge whether machines can do this type of thinking. For years, he has contributed major research to the field of deep learning, including at Google, where he worked as a software engineer from 2015 until this past November; he wants generative AI to be revolutionary, but worries that the industry has strayed. In 2019, Chollet created the Abstraction and Reasoning Corpus for Artificial General Intelligence, or ARC-AGI—an exam designed to show the gulf between AI models’ memorized answers and the “fluid intelligence” that people have. Drawing from cognitive science, Chollet described such intelligence as the ability to quickly acquire skills and solve unfamiliar problems from first principles, rather than just memorizing enormous amounts of training data and regurgitating information. (Last year, he launched the ARC Prize, a competition to beat his benchmark with a $1 million prize fund.)
You, a human, would likely pass this exam. But for years, chatbots had a miserable time with it. Most people, despite having never encountered ARC-AGI before, get scores of roughly 60 to 70 percent. GPT-3, the program that became ChatGPT, the legendary, reality-distorting bot, scored a zero. Only recently have the bots started to catch up.
How could such powerful tools fail the test so spectacularly for so long? This is where Chollet’s definition of intelligence comes in. To him, a chatbot that has analyzed zillions of SAT-style questions, legal briefs, and lines of code is not smart so much as well prepared—for the SAT, a law-school exam, advanced coding problems, whatever. A child figuring out tricky word problems after just learning how to multiply and divide, meanwhile, is smart.
ARC-AGI is simple, but it demands a keen sense of perception and, in some sense, judgment. It consists of a series of incomplete grids that the test-taker must color in based on the rules they deduce from a few examples; one might, for instance, see a sequence of images and observe that a blue tile is always surrounded by orange tiles, then complete the next picture accordingly. It’s not so different from paint by numbers.
The test has long seemed intractable to major AI companies. GPT-4, which OpenAI boasted in 2023 had “advanced reasoning capabilities,” didn’t do much better than the zero percent earned by its predecessor. A year later, GPT-4o, which the start-up marketed as displaying “text, reasoning, and coding intelligence,” achieved only 5 percent. Gemini 1.5 and Claude 3.7, flagship models from Google and Anthropic, achieved 5 and 14 percent, respectively. These models may have gotten lucky on a few puzzles, but to Chollet they hadn’t evinced a shred of abstract reasoning. “If you were not intelligent, like the entire GPT series,” he told me, “you would score basically zero.” In his view, the tech barons were not even on the right path to building their artificial Einstein.
Chollet designed the grids to be highly distinctive, so that similar puzzles or relevant information couldn’t inadvertently be included in a model’s training data—a common problem with AI benchmarks. A test taker must start anew with each puzzle, applying basic notions of counting and geometry. Most other AI evaluations and standardized tests are crude by comparison—they aren’t designed to evaluate a distinct, qualitative aspect of thinking. But ARC-AGI checks for the ability to “take concepts you know and apply them to new situations very efficiently,” Melanie Mitchell, an AI researcher at the Santa Fe Institute, told me.
To improve their performance, Silicon Valley needed to change its approach. Scaling AI—building bigger models with more computing power and more training data—clearly wasn’t helping. OpenAI was first to market with a model that even came close to the right kind of problem-solving. The firm announced a so-called reasoning model, o1, this past fall that Altman later called “the smartest model in the world.” Mark Chen, OpenAI’s chief research officer, told me the program represented a “new paradigm.” The model was designed to check and revise its approach to any question and to spend more time on harder ones, as a human might. An early version of o1 scored 18 percent on ARC-AGI—a definite improvement, but still well below human performance. A later iteration of o1 hit 32 percent. OpenAI was still “a long way off” from fluid intelligence, Chollet told me in September.
That was about to change. In late December, OpenAI previewed a more advanced reasoning model, o3, that scored a shocking 87 percent on ARC-AGI—making it the first AI to match human performance on the test and the best-performing model by far. Chollet described the program as a “genuine breakthrough.” o3 appeared able to combine different strategies on the fly, precisely the kind of adaptation and experimentation needed to succeed on ARC-AGI.
Unbeknownst to Chollet, OpenAI had kept track of his test “for quite a while,” Chen told me in January. Chen praised the “genius of ARC,” calling its resistance to memorized answers a good “way to test generalization, which we see as closely linked to reasoning.” And as the start-up’s reasoning models kept improving, ARC-AGI resurfaced as a meaningful challenge—so much so that the ARC Prize team collaborated with OpenAI for o3’s announcement, during which Altman congratulated them on “making such a great benchmark.”
Chollet, for his part, told me he feels “pretty vindicated.” Major AI labs were adopting, even standardizing, his years-old ideas about fluid intelligence. It is not enough for AI models to memorize information: They must reason and adapt. Companies “say they have no interest in the benchmark, because they are bad at it,” Chollet said. “The moment they’re good at it, they will love it.”
Many AI proponents were quick to declare victory when o3 passed Chollet’s test. “AGI has been achieved in 2024,” one start-up founder wrote on X. Altman wrote in a blog post that “we are now confident we know how to build AGI as we have traditionally understood it.” Since then, Google, Anthropic, xAI, and DeepSeek have launched their own “reasoning” models, and the CEO of Anthropic, Dario Amodei, has said that artificial general intelligence could arrive within a couple of years.
But Chollet, ever the skeptic, wasn’t sold. Sure, AGI might be getting closer, he told me—but only in the sense that it had previously been “infinitely” far away. And just as this hurdle was cleared, he decided to raise another.
Last week, the ARC Prize team released an updated test, called ARC-AGI-2, and it appears to have sent the AIs back to the drawing board. The full o3 model has not yet been tested, but a version of o1 dropped from 32 percent on the original puzzles to just 3 percent on the new version, and a “mini” version of o3 currently available to the public dropped from roughly 30 percent to below 2 percent. (An OpenAI spokesperson declined to say whether the company plans to run the benchmark with o3.) Other flagship models from OpenAI, Anthropic, and Google have achieved roughly 1 percent, if not lower. Human testers average about 60 percent.
If ARC-AGI-1 was a binary test for whether a model had any fluid intelligence, Chollet told me last month, the second version aims to measure just how savvy an AI is. Chollet has been designing these new puzzles since 2022; they are, in essence, much harder versions of the originals. Many of the answers to ARC-AGI were immediately recognizable to humans, while on ARC-AGI-2, people took an average of five minutes to find the solution. Chollet believes the way to get better on ARC-AGI-2 is to be smarter, not to study harder—a challenge that may help push the AI industry to new breakthroughs. He is turning the ARC Prize into a nonprofit dedicated to designing new benchmarks to guide the technology’s progress, and is already working on ARC-AGI-3.
Reasoning models take bizarre and inhuman approaches to solving these grids, and increased “thinking” time will come at substantial cost. To hit 87 percent on the original ARC-AGI test, o3 spent roughly 14 minutes per puzzle and, by my calculations, may have required hundreds of thousands of dollars in computing and electricity; the bot came up with more than 1,000 possible answers per grid before selecting a final submission. Mitchell, the AI researcher, said this approach suggests some degree of trial and error rather than efficient, abstract reasoning. Chollet views this inefficiency as a fatal flaw, but corporate AI labs do not. If chatbots achieve fluid intelligence in this way, it will not be because the technology approximates the human mind: You can’t just stuff more brain cells into a person’s skull, but you can give a chatbot more computer chips.
In the meantime, OpenAI is “shifting towards evaluations that reflect utility as well,” Chen told me, such as tests of an AI model’s ability to navigate and take actions on the web—which will help the company make better, although not necessarily smarter, products. OpenAI itself, not some third-party test, will ultimately decide when its products are useful, how to price them (perhaps $20,000 a year for a “Phd-level” bot, according to one report), and whether they’ve achieved AGI. Indeed, the company may already have its own key AGI metric, of a sort: As The Information reported late last year, Microsoft and OpenAI have come to an agreement defining AGI as software capable of generating roughly $100 billion in profits. According to documents OpenAI distributed to investors, that determination “is in the ‘reasonable discretion’ of the board of OpenAI.”
And there’s the problem: Nobody agrees on what’s being measured, or why. If AI programs are bad at Chollet’s test, maybe it just means that they have a hard time visualizing colorful grids rather than anything deeper. And bots that never solve ARC-AGI-2 could generate $100 billion in profits some day. Any specific test—the LSAT or ARC-AGI or a coding puzzle—will inherently contradict the notion of general intelligence; the term’s defining trait may be its undefinability.
The deeper issue, perhaps, is that human intelligence is poorly understood, and gauging it is an infamously hard and prejudiced task. People have knacks for different things, or might arrive at the same result—the answer to a math problem, the solution to an ARC-AGI grid—via very different routes. A person who scores 30 percent on ARC-AGI-2 is in no sense inferior to someone who scores 90 percent. The collision of those differing routes and minds is what sparks debate, creativity, and beauty. Intentions, emotions, and lived experiences drive people as much as any logical reasoning.
Human cognitive diversity, in other words, is a glorious jumble. How do you even begin to construct an artificial version of that? And when that diversity is already so abundant, do you really want to?
After a few months of shivering through Severance’s blank white corridors and icy exterior shots, I’ve appreciated the sultry visual texture of The White Lotus’s third season: the vivid prints of high-end resort wear; the ominous blue of the ocean; the verdant setting (as wild and seething as anything manicured into luxury-hotel perfection can be). The show is thrilling as a sensory experience, humming with sinister percussive beats and the occasional muffled animal squawk in the distance. Against this backdrop, it feels only natural that we’d fall in love with the characters who seem the most real, the most alive.
I’m talking, of course, about Chelsea, played by Aimee Lou Wood, and Chloe, played by Charlotte Le Bon—two gorgeous women who meet at a bar after Chelsea says, “I love your outfit,” and Chloe replies, “Thank you! I love your teeth.” This quick moment set off a good-natured riot of online debate—labeled the “smile discourse” by Allure—about what it means to see not just imperfect teeth on-screen, but also imperfect teeth on women who are undeniable knockouts. I’ll defer to others regarding the particulars of dental trends, but I can tell you how it made me feel to see such gloriously irregular beauty amid all the identical Instagram faces with the same Tic-Tac veneers, stenciled eyebrows, and contoured cheekbones: relieved.
Chelsea, played by Aimee Lou Wood, on The White Lotus (Fabio Lovino / HBO)
Lately, I’ve been finding myself more and more unsettled by digital faces tweaked and pixelated into odd perfection and real bodies buffed and whittled down into obscene angularity—women who look less like flesh-and-blood beings than porcelain ornaments. At the Oscars last month, Rachel Tashjian wrote in The Washington Post, the eerie flawlessness of so many red-carpet looks seemed to encapsulate “how weight loss drugs and technology, including photo editing and AI-generated imagery, have ushered in an outrageous drive for perfection that has overtaken Hollywood.” If you compare the poreless, rose-toned face of the superstar Ariana Grande with the sculpted cheekbones and button nose of the Spanish influencer Aitana Lopez, it’s hard to discern even infinitesimally minute flaws in either. Unlike Grande, though, Lopez is computer-generated—one of a new breed of models with hundreds of thousands of followers and horny men continually sliding into her DMs, despite the fact that she’s wholly nonexistent.
Much has been written over the past few months about the propagandist tendencies of artificially generated art—the way it’s been gleefully adopted by right-wing trolls to create photorealistic but recognizably fake images of Elon Musk giving out wads of cash, or the surreal 30-second clip that Donald Trump recently posted imagining Gaza as a gilded beachside temple to wealth and potentates. These kinds of pictures are intended to provoke—to catch the eye with their mawkish absurdity and uncanny-valley optics. But to me at least, the beautified AI faces are no less offensive. They reflect back at us toxic values that we’re in thrall to, and capture none of the qualities we should truly appreciate. The writer Daphne Merkin once observed that in reality, we find imperfection enchanting because we recognize “that behind the visceral image lies an internal life.” Which, I’d wager, is why the wonky smiles of Wood and Le Bon are so compelling in this moment: They assert the intangible beauty of having a soul.
We have never, as mere human bags of flesh and bone, been so perfectible. We’ve never had as many tools in our arsenal with which to maximize our superficial value: weight-loss drugs that can make slim bodies even smaller, Botox and fillers that smooth out wrinkles, contouring pens that define features. This is even before we get into the realm of augmented reality. On TikTok, I can broadcast myself using a filter that makes me look exactly as I did at 23: lifted, smoothed, softer, and also somehow lighter and less harried. Ninety percent of British women and nonbinary people polled in 2020 confessed to sometimes using filters before posting selfies, and 85 percent to using external editing software such as FaceTune to tweak pictures of themselves. Every single woman surveyed said they had been served videos promoting plastic surgery in their feeds: before-and-after reels selling lip fillers, teeth-whitening treatments, butt enhancers. A few months ago, I too was suddenly inundated with clips of scrub-wearing surgeons “analyzing” Lindsay Lohan’s face, after new images of the actor suddenly began to circulate revealing catlike eyes, a heart-shaped face, and the skin of a well-rested teenager.
What struck me about the Lohan images was less what work she had or hadn’t done, and more the way in which, virtually overnight, a battalion of influencer-doctors jumped onto social media, selling us on the idea of our own transformation. To some extent, each generation has lived through its own freakout regarding what technological advances might be doing to beauty standards, and to our fragile sense of self. In 2006, The Guardiannoted that Photoshop was making even supermodels outraged, and that tabloids were reacting to the prevalence of perfected images by seeking out unflattering candid shots for balance: stars with straggly hair, or visible cellulite, or slight paunches. In 2019, the cultural critic Jia Tolentino coined the term Instagram Face for the “single, cyborgian look” being popularized on social media by models and influencers. And in her new book, Searches: Selfhood in the Digital Age, the journalist Vauhini Vara writes about how technology has managed to change the way human beings look by altering our ideals, giving us a funhouse-mirror reflection of how we think we should look. “To live like this, endlessly comparing our imperfect fleshy selves with the sanitized digital simulacra of selfhood that appears online and finding ourselves wanting,” Vara notes, “exerts such a subtle psychic violence that we might not even be aware of it as it’s happening.”
In some ways, though, technology also primed us for what was to come. The more fault we’re compelled to find with our own unsymmetrical, lined, irredeemably lived-in faces, the more we’re set up to be swayed by the unreal smoothness of AI imagery. In 2023, when the AI image generator Stable Diffusion XL was launched, the company behind it boasted that the product created the most photorealistic images yet available. (It offered, by way of emphasis, a picture of a panda in a spacesuit sitting at a bar.) What was clear early on, though, was that Stable Diffusion XL had the same biases and prejudices humans do, amplified to an absurd extent. Prompts for “a person at social services” generated pictures of predominantly Black women; prompts for “a productive person” generated largely white men in suits. AI image generators also had, as my former colleague Caroline Mimbs Nyce reported, a “hotness problem,” generating pictures of people who were all improbably attractive. Possibly this is because they were built by scanning edited and airbrushed photos—not just of professionally attractive people, but of us. (Every time you FaceTune a selfie, the theory goes, a neural network further distorts its sense of what humans actually look like.)
Recently, I asked Microsoft’s Image Creator for a picture of a normal woman. It gave me four extraordinarily beautiful women with curly hair, sculpted jawlines, and plump lips. (All four were wearing glasses, a supposed de-beautifying trick that didn’t work in She’s All That and doesn’t work now.) Then I asked for a picture of an average woman, for which I received four images of radiantly smiling women in baggy sweaters with slightly frizzy hair. Finally, prompted to give me a picture of an average 42-year-old woman (my birthday is this month), the program gave me the eeriest images of all: four Anne Hathaway look-alikes with monstrously oversize grins and visible clavicles, betraying only slight lines around their eyes, and inexplicably surrounded by other grinning hot people, as if advertising a cult.
What’s so unsettling about these images, I think, is how they reflect what we’re allowing technology to do to us, what it’s already done. Given the ability to amend our own faces, we’ve helped normalize and propagate a horribly restrictive vision of beauty and humankind, and the more we distort ourselves in turn, the more confining the ideal becomes. Recently, the art historian Sonja Drimmer argued that artificial intelligence was “essentially useless” for the purpose of studying history, because historians “look for untold stories” and “elements of the history of mankind that are novel and unexpected.” Programs such as ChatGPT, by contrast, can only skim and interpret texts and images that already exist, extrapolating them into likely outcomes. If you’re looking for nuance, or uncertainty, or subtext, it can’t help you.
With regards to beauty, I’d bargain that everyone knows someone who shouldn’t, by all superficial accounts, be attractive, and yet they are. Because: We’re better than computers at reading between the lines and can see other people’s faces not just as structural compositions of bone and skin, but also as reflections of personality, of humanity, of depth. And the more we can defend beauty as nonconformist, as the essence of something internal and unmeasurable, the more we protect ourselves from the narrowing grip of techno-homogenization. In The White Lotus, and in reality, Wood’s face isn’t just beautiful. It’s guileless, openhearted, kind, tender. “You’re never going to look like what you think perfect is,” the actor told Glamour. And the more I see perfect, the less I can bear it.
When you buy a book using a link on this page, we receive a commission. Thank you for supporting The Atlantic.
Yesterday afternoon, Donald Trump celebrated America’s so-called Liberation Day by announcing a slew of tariffs on dozens of countries. His plan, if fully implemented, will return the United States to the highest tariff duty as a share of the economy since the late 1800s, before the invention of the automobile, aspirin, and the incandescent light bulb. Michael Cembalest, the widely read analyst at JP Morgan Wealth Management, wrote that the White House announcement “borders on twilight zone territory.”
The most fitting analysis for this moment, however, does not come from an economist or a financial researcher. It comes from the screenwriter William Goldman, who pithily captured his industry’s lack of foresight with one of the most famous aphorisms in Hollywood history: “Nobody knows anything.”
You’re not going to find a better three-word summary of the Trump tariffs than that. If there’s anything worse than an economic plan that attempts to revive the 19th-century protectionist U.S. economy, it’s the fact that the people responsible for explaining and implementing it don’t seem to have any idea what they’re doing, or why.
On one side, you have the longtime Trump aide Peter Navarro, who has said that Trump’s tariffs will raise $6 trillion over the next decade, making it the largest tax increase in American history. On another, you have pro-Trump tech folks, such as Palmer Luckey, who have instead claimed that the goal is the opposite: a world of fully free trade, as countries remove their existing trade barriers in the face of the new penalties. On yet another track, there is Stephen Miran, the chair of the Council of Economic Advisers, who has suggested that the tariff salvo is part of a master plan to rebalance America’s relationship with the global economy by reducing the value of the dollar and reviving manufacturing employment in the United States.
These three alleged goals—raising revenue, restoring free trade, and rejiggering the global economy—are incompatible with one another. The first and second explanations are mutually exclusive: The state can’t raise tax revenue in the long run with a levy that is designed to disappear. The second and third explanations are mutually exclusive too: You can’t reindustrialize by doubling down on the global-trade free-for-all that supposedly immiserated the Rust Belt in the first place. Either global free trade is an economic Valhalla worth fighting for, or it’s the cursed political order that we’re trying desperately to destroy.
As for Trump’s alleged devotion to bringing back manufacturing jobs, the administration has attacked the implementation of the CHIPS bill, which invested in the very same high-tech semiconductors that a strategic reindustrialization effort would seek to prioritize. There is no single coherent explanation for the tariffs, only competing hypotheses that violate one another’s internal logic because, when it comes to explaining this economic policy, nobody knows anything.
One might expect clarity from Treasury Secretary Scott Bessent. But even he doesn’t seem to understand what’s going on. The “tariff gun will always be loaded and on the table, but rarely discharged,” he said last year. So much for that. Yesterday, a Bloomberg reporter asked Bessent if the Trump administration has plans to negotiate with America’s trading partners. “We’re just going to have to wait and see,” he said. Was the administration ready to negotiate with the European Union, China, or India? “We’ll see.” Asked why Canada and Mexico were missing from the president’s list of tariffs, he switched it up: “I’m not sure.” Nobody knows anything.
By the numbers, the tariffs are less an expression of economic theory and more a Dadaist art piece about the meaninglessness of expertise. The Trump administration slapped 10 percent tariffs on Heard Island and McDonalds Islands, which are uninhabited, and on the British Indian Ocean Territory, whose residents are mostly American and British military service members. One of the highest tariff rates, 50 percent, was imposed on the African nation of Lesotho, whose average citizen earns less than $5 a day. Why? Because the administration’s formula for supposedly “reciprocal” tariff rates apparently has nothing to do with tariffs. The Trump team seems to have calculated each penalty by dividing the U.S. trade deficit with a given country by how much the U.S. imports from it and then doing a rough adjustment. Because Lesotho’s citizens are too poor to afford most U.S. exports, while the U.S. imports $237 million in diamonds and other goods from the small landlocked nation, we have reserved close to our highest-possible tariff rate for one of the world’s poorest countries. The notion that taxing Lesotho gemstones is necessary for the U.S. to add steel jobs in Ohio is so absurd that I briefly lost consciousness in the middle of writing this sentence.
If the tariffs violate their own internal logic and basic common sense, what are they? Most likely, they represent little more than the all-of-government metastasis of Trump’s personality, which sees grandiosity as a strategy to pull counterparties to the negotiating table and strike deals that benefit Trump’s ego or wallet. This personality style is clear, and it has been clearly stated, even if its application to geopolitics is confounding to observe. “My style of deal-making is pretty simple and straightforward,” Trump writes in The Art of the Deal. “I aim very high, and then I just keep pushing and pushing and pushing to get what I’m after. Sometimes I settle for less than I sought but in most cases I still end up with what I want.”
One can see this playbook—threat, leverage, concession, repeat—playing out across all of society. It’s happening in trade. It’s happening in law. It’s happening in academia. In the first two months of his second term, Trump has already squeezed enormous concessions out of white-shoe law firms and major universities. Trump appears to care more about the process of gaining leverage over others—including other countries—than he does about any particular effective tariff rate. The endgame here is that there is no endgame, only the infinite game of power and leverage.
Trump’s defenders praise the president for using chaos to shake up broken systems. But they fail to see the downside of uncertainty. Is a textile company really supposed to open a U.S. factory when our trade policy seems likely to change every month as Trump personally negotiates with the entire planet? Are manufacturing firms really supposed to invest in expensive factory expansions when the Liberation Day tariffs caused a global sell-off that signals an international downturn? Trump’s personality is, and has always been, zero-sum and urgent, craving chaos, but economic growth is positive-sum and long-term-oriented, craving certainty for its largest investments. The scariest thing about the Trump tariffs isn’t the numbers, but the underlying message. We’re all living inside the president’s head, and nobody knows anything.
