My friend Thomas and I text each other music. We do it on a janky WhatsApp feed that deletes our messages every few days. I listen to it on weeknights, after the kids are asleep, while I wash the dishes really slowly.
On a good day, Thomas might send me Kendrick, and I might respond with Karol G. On a great day, he sends me a video of a Mardi Gras Indian in full regalia trash-singing his rival krewes over a slick New Orleans bounce beat. Then there was the night last February when he sent me a YouTube link to a song by an artist from Kerala named Hanumankind. It was called “Big Dawgs.”
I clicked on the video. A voiceover warns against trying these stunts at home. Someone screams at the voice to “shut the f— up.” A motorcycle roars. The camera zooms in on Hanumankind rapping on the dirt floor of a spherical wooden amphitheater. Picture Mad Max’s Thunderdome, except it’s made not of steel but rickety four by fours.
Motorcycles shoot into frame … on the wall. The beat kicks in. The ground shakes. An old white sedan shows up on the floor. Now the car is driving on the wall. Rolling through the city with the big dawgs, f— the law, lawyer with me, we don’t gotta call, raps Hanumankind, his body hanging out the passenger window.
I put the sponge down. A woman in white slacks and a nice blue dress enters the amphitheater. She looks like she’s going to a garden party. Now she gets on a motorcycle and starts tearing across the wall. At school, I used to fight the bullies. Now I’m fighting with the law, I guess some things don’t leave you fully.
Somehow, every ten seconds, the video goes harder. Now the lady in white slacks is in the car. The car is now on the wall. Now she is on the car—on top of the car, driving on the side of the wall, bopping her head like there’s a nice tune on the radio.
The dishes will wait. I restart the video. Then I do it again, and again, and again.
The Argentine poet and essayist Jorge Luis Borges wrote about the Zahir, an object that would inexplicably captivate its beholder and eventually drive them mad. I don’t know about all that, but I do know that every one of my text feeds got a link to Big Dawgs. Who am I kidding? Every in-person conversation ended with me taking out my phone and sending that person a link to Big Dawgs.
About a week later, Thomas texted again.
“I’m obsessed with Big Dawgs and I don’t know why,” I blurted.
“Me too,” he shot back. “But I know why,” he said. “There is zero AI in that video. It’s just real people doing wildly dangerous things. And pulling it off.”
He was right. The video was so brazen, so joyous, so defiant. AI would have flattened it. There’s no death to defy if you’re an AI.
We didn’t know it, but as we texted each other, thousands of American actors who are paid to do those kinds of stunts had been on strike for seven months. Their union, the Screen Actors Guild, had walked out of their jobs—to protect them against AI.
“I stick to walls. I beat people up. I get beaten up constantly, get electrocuted, and turn invisible,” said Jasiri Brooks, a Guild member and motion capture performer for the Spider-Man: Miles Morales video game. He did it gladly. But he feared his employers would use AI to purloin his movements and add his stunts into future games—and cut him, the real Jasiri Brooks, out of it. “[W]e’re saying at the very least, please inform us and allow us to consent to the performances that you are generating with our AI doubles.”
For half a century, there was an institution to protect us from private-sector attempts to surveil, control, and profit from our talents: Congress. From 1970 to 2009, Congress passed almost a dozen laws that made it harder for corporations to obtain, use, and monetize information about our private and professional lives.
AI companies are shattering the social contract—and making billions doing it.
Now, AI companies are trying to digitize and monetize what makes us human: The way we move, speak, and interact; the way we heal each other through empathy and trust. Labor unions see AI as a signal threat to their members’ livelihoods. Meanwhile, Congress has done nothing to protect American workers—no privacy laws, no measures to protect jobs or to rein in the power of AI companies. Nothing.
While people speculate about the scope or permanence of AI-driven job losses, I can’t stop thinking about the glaring disparity in the ambitions of the AI class and the elected leaders who are supposed to protect us. AI companies aren’t renegotiating the social contract; they’re shattering it—and making billions doing it. Meanwhile, people in Congress hold roundtables to discuss “guardrails” and “guidelines” to “rein in” AI. Watching this, I have a nagging feeling that maybe we should stop listening to the people in Congress and start listening to the workers whose job it is to defy gravity. They’re the only ones who seem to understand the assignment.
Ironically, when I got Thomas’ link, I was “the law.” I was a commissioner at the Federal Trade Commission, the nation’s consumer protection and anti-monopoly watchdog. But when the president illegally fired me, we weren’t messing with guys like Hanumankind. We were publicly probing how Amazon, Microsoft, and OpenAI might have daisy-chained their AI investments to tie their companies together while avoiding scrutiny from antitrust enforcers. We also investigated allegations that Snapchat’s AI chatbot hurt the kids it interacted with, and publicly revealed that when we referred a complaint against the company to the Department of Justice.
But when you have that kind of job, you don’t just investigate and sue billionaires. You also try to understand how they move in the market.
Over time, I noticed something.
On the surface, the AI oligarchs look quite different from each other. On the one hand, you have OpenAI CEO Sam Altman, whose dead-eyed TV interviews make Mark Zuckerberg look like Zohran Mamdani. On the other, you have the cerebral Dario Amodei, CEO of Anthropic, who feels kind of bad that he’s going to replace you and your kids’ jobs with lines of code that periodically just make shit up.
But in their actions, the CEOs all agree on this: There is nothing—not one thing—they are not entitled to.
Can they non-consensually scan countless books from countless libraries, digitizing and monetizing an astonishing fraction of human knowledge, despite something called “copyright”?
Yes.
Can they build vast data centers to fuel their creations, burdening a small city with the electricity demands of a second, entirely separate small city, and make the people in the actual city subsidize the data center through higher rates and bills?
Yes.
Can they watch teens lose years of their lives scrolling Instagram and TikTok, watch mental health problems and suicide attempts skyrocket… and then blithely plan new AI chatbots to lure the next generation of lonely people through “sensual” conversations with children or actual “erotica”?
In his most intimate anecdote, Amodei claims that AI systems can replace humans in professions that require the utmost trust and empathy.
Yes indeed. But the acme of their ambition is what they want to do with labor.
The clearest statement of that vision is in a 19,000-word monograph, “The Adolescence of Technology,” self-published by Mr. Amodei on his personal blog. On its face, the essay is a beneficent, step-by-step treatise on how our government should confront the rogue bioweapons, “billions of fully automated armed drones,” mass surveillance, and mass unemployment made possible by advanced AI systems … like those made by Anthropic.
It is an arsonist’s guide to firefighting. In that sense, what makes the document interesting is not Mr. Amodei’s policy recommendations but rather his predictions about how the fire will burn.
When he turns to labor, Amodei first expounds on his earlier prediction that AI would “displace” half of all white-collar jobs in the next one to five years. Ultimately, the full breadth of his plans is made explicit: “In the end, AI will be able to do everything, and we need to grapple with that.”
In his most intimate anecdote, Amodei claims that AI systems can replace humans in professions that require the utmost trust and empathy:
AI is already widely used for customer service. Many people report that it is easier to talk to AI about their personal problems than to talk to a therapist—that the AI is more patient. When my sister was struggling with medical problems during a pregnancy, she felt she wasn’t getting the answers or support she needed from her care providers, and she found Claude to have a better bedside manner (as well as succeeding better at diagnosing the problem). I’m sure there are some tasks for which a human touch really is important, but I’m not sure how many…
Computers compute. Amodei wants you to think they can emote. Or rather, he wants you to think computers can mimic human actions or emotions so effectively that you forget that this is impossible.
It isn’t just Amodei and Anthropic. A range of AI companies are actively trying to take those jobs from humans. For them, it is a burgeoning market waiting to be tapped. For the people who depend on those jobs, it is a catastrophe.
Start with customer service, which Amodei notes is already being remade by AI.
For years, customer service call centers have offered high school graduates a chance at a middle-class income. Often driven by commissions, the centers reward agents who can calm or charm their way into a sale—or turn an angry customer into a satisfied one.
Now, those workers are being replaced with AI by the thousands. And they have been trained on the conversations of the workers they’re being used to replace. How did that happen?
Constant, minute surveillance.
The modern call center worker is videotaped, their calls are recorded, and their keystrokes are logged. The words they use, their pace, and the precise tone they use when saying those words are measured and analyzed in real time. (One program assigns people “overall empathy scores,” defined as “100 * [(Number of empathetic phrases – Number of unhelpful phrases) / (Total number of empathetic and unhelpful phrases)].”) And when something goes wrong, an AI agent will provide them immediate “guidance” to “elevat[e] their performance by offering next best action suggestions.”
A Cornell University survey found that the average call center uses five forms of surveillance on its employees.
The companies say this surveillance will unleash a new wave of productivity. The workers find it punitive and baffling.
One program triggered an alarm for emotional distress whenever a customer started laughing. Another was told by her (human) supervisor that she “was doing a really great job,” but kept getting dinged by AI’s low “sentiment” scores on her calls. She thought she wasn’t using the right words with callers, but no one would tell her the words she needed to say.
Still another AI chatbot confidently told a seasoned operator that his next caller wanted a home repair plan, only for that operator to discover that the enraged caller had a repair plan, his house was flooding, and no one was there to repair it.
It’s like having a “computer… standing over your shoulder and arbitrarily deciding whether you get to keep your job or not,” said one worker. Another recalled, “the stress made me sick to my stomach and unable to get out of bed in the morning to do my job.” A separate Cornell study found that most call center workers had been prescribed medication for a stress- or anxiety-related illness, and that one in four took it “constantly.”
The companies say this surveillance will unleash a new wave of productivity. The workers find it punitive and baffling.
But we expected that AI would quickly come for customer service jobs; analysts predicted that. Jobs like those of the actors in Big Dawgs and the American motion-capture performers like Jasiri Brooks were supposedly a rung or two lower on the AI chopping block.
Yet the attempt to digitize and replace their jobs occurred quite quickly as well, and came to a head well before video game actors like Brooks went on strike in July 2024.
A year before that, screen actors called their own strike against the film and TV studios (not the video game companies that Brooks worked for). Like Brooks, the screen actors feared the studios would digitize their likenesses and, unbeknownst to them, use AI to add replicas of them to future productions—concerns that were validated when the studios made a proposal to govern the scanning of background actors, also known as “extras.” According to the Guild’s chief negotiator, Duncan Crabtree-Ireland, the studios demanded that extras “should be able to be scanned, get one day’s pay, and their companies should own that scan… and should be able to use it for the rest of eternity on any project they want, with no consent and no compensation.”
Suddenly, the press, social media, and public protests were flooded with bizarre accounts of digital scans. Background actors were being told to report to specialized rooms or tents. Upon entering, they would be subject to 360-degree head-to-toe scans by hundreds of cameras. If the actors asked what the scans were for, some were told they were for special effects, toys, or “whatever they needed.” Many heard that their likeness could be used in “perpetuity.” Some were given no answers. When the actors asked whether they would be compensated for the scans, some received $25, while others received nothing.
A few actors were shown digital replicas of themselves “walking around in the background on screen” in scenes they’d never acted in. If they protested, they’d be blackballed from future productions.
The concerns weren’t limited to extras. Brian Cox, the Shakespearean actor who played the patriarch in Succession, told protesters about how, during a recent television appearance, he “was handed a list of things that artificially intelligent Brian Cox was going to say.”
“Artificially intelligent Brian Cox was going to do animal impersonations. I’ve never done a fucking animal impersonation in my life,” Cox said. “This is going to happen to everybody. Nobody is exempt from this.”
Then, of course, there are the AI companies trying to replace mental health workers.
Last month, mental health care workers at Kaiser Permanente went on a 24-hour strike, stopping work at the company’s facilities in Northern California. If you are in a mental health crisis in Oakland, San Francisco, or Sacramento—and you have Kaiser as your insurer—these are the folks who answer when you call for help.
Like the movement actors and the screen actors, their bosses were swapping them out for an algorithm.
During the strike, I sat in a meeting of the California Labor Federation at the Sheraton Grand Hotel in Sacramento. I listened to a licensed clinical social worker sound off on why they had to do it.
Ilana Marcucci-Morris is a member of the National Union of Health Care Workers. Her arms and neck are prominently tattooed. And she had spent years running 15-minute triage interviews on Kaiser’s mental health line, figuring out who needed appointments right away and who could wait a little longer. Doing that work, she explained, you develop an instinct for distinguishing the college student freaking out about an upcoming exam and the high-risk individual who has access to a gun, who has previous suicide attempts, or who might be contemplating a school shooting.
Who’s doing the work now? “It’s a robot instead,” she said.
More specifically, according to a complaint filed by the NUHCW, the workers alleged that triage calls were now being handled by “unlicensed, untrained clerical staff” in under five minutes. They were using rote questionnaires that they fed into a software program called LUCET, which generated an “acuity score” to guide licensed health care workers in scheduling patients. But no, the complaint says: “In most cases, the triage decisions are not reviewed by licensed and trained therapists.”
As a result, the workers alleged, the system might not schedule urgent cases first, but last. Or, it might miss that a caller has addiction issues that require specialized treatment, scheduling them for a standard therapy intake ten days later, where an actual clinician realizes the mistake and must refer the patient for a new, separate consult with addiction services.
Now, Marcucci-Morris told The Guardian, by the time she sees patients, they’re often in far worse shape than they were before. “Thank God they’re still alive,” she said.
If this essay were a thread on social media, this is when someone would comment with the meme of a cartoon dog sipping coffee while his house goes up in flames. “This is fine,” he says.
It is not fine. But we don’t live in that house. Or at least, we do not have to.
We could have a future where AI dominates workers: Where it awkwardly approximates their grace and empathy enough to replace them (at least for now); where it browbeats the remaining workers and overrides their judgment; and where a few men get the rat’s share of the profits.
The few meaningful bulwarks against AI at work were not passed by Congress, nor were they won by lawyers filing lawsuits: They were won at the bargaining table by organized labor. And I say this as a lawyer who writes bills and files lawsuits.
But we could also live in a world where AI augments workers: Where their charm and artistry are theirs to share or keep, where AI automates rote tasks and lets humans focus on the hard judgment calls that only a human should make, and where workers earn their fair share from the fruits of these innovations.
I know that world is possible because I see flashes of it, thanks to those workers.
Right now, movement actors in Hollywood told, in detail, how their movements will be used; they can choose to allow or prohibit those uses, and if they allow them, they will be paid for that. Right now, screen actors must be given the same set of choices, and they, too, must be paid. Right now, in some of the country’s biggest customer service call centers, workers cannot be disciplined or fired because an AI system said so. Those decisions can only be made by humans.
Why? Because the video game actors won their strike. The screen actors did, too. And in each of those call centers, those protections were won, company by company, in contract negotiations by the Communications Workers of America.
The media tirelessly covers Capitol Hill. Yet the few bright bulwarks against AI at work were not passed by Congress, nor were they won by lawyers filing lawsuits: They were won at the bargaining table by organized labor. And I say this as a lawyer who writes bills and files lawsuits.
Indeed, it is in state legislatures—far from Congress and in states where union membership is strongest—where the clearest picture of that second world comes into view.
In that hotel conference hall in Sacramento, on the day Ms. Marcucci-Morris warned of the crisis lines staffed by an intern and an algorithm, I watched California’s teachers, nurses, bus and truck drivers, drywall hangers, and electricians plan to build that world, one bill at a time.
The laws they were preparing to advocate for in California’s capital would prohibit employers from using the most invasive and exploitative forms of AI surveillance at work, ban those bosses from forcing people to train their AI replacements, ban driverless trucks careening through crowded public streets, and ensure that health care decisions were always made not by robots, but by humans. The bill with the most support in the room would give state law enforcers more power to break up the corporate monopolies that somehow convinced themselves that these practices were acceptable in the first place.
Unlike their counterparts in Congress, California legislators have a real shot at enacting these laws. Some bills are well on their way to passage. In fact, last year, the California legislature passed not one but two laws to protect actors’ digital replicas within two months of the start of the video game actors’ strike. (One bill governed the replicas of deceased actors; the other invalidated certain contracts where actors signed away their rights with little information or without a lawyer or labor representative present.)
I recently watched a video of United Auto Workers President Shawn Fain testify before Congress, calling for a 32-hour workweek. Fain noted that in 1933, the U.S. Senate debated a 30-hour workweek. The men who debated that proposal were convinced that, while the “improvement of machinery” had allowed one factory worker to “do the work of 60 men,” the profit from that efficiency was not going to the worker: “It has been going to the man who owned the machine.”
The 1933 Senate passed the 30-hour workweek bill, but the industry’s attacks defeated it. At the urging of organized labor, Franklin Delano Roosevelt codified a 40-hour workweek in 1940.
Since then, digital machinery has brought about “a 400 percent increase in productivity,” said Fain. “But nothing’s changed”—the money is again going to the men who own the machines. And the 40-hour workweek remains in place. So the United Auto Workers made a 32-hour workweek one of the key demands in their 2024 strikes.
Many social, political, and economic movements have shaped our country. Those movements gave some people a degree of liberty unthinkable at our founding; they also gave corporate America more power than almost any flesh-and-blood human. Yet of all those movements, only one changed the basic structure of time in modern society. Only one had the gall to imagine it possible and then make it a reality.
Maybe we should listen to them.
Thomas hasn’t managed to top Big Dawgs. Neither have I. But I do have a story about a different video that did use AI.
My wife, Sima, and I do Elf-on-a-Shelf for Christmas. More accurately, Sima does Elf-on-a-Shelf every night while I grumble that I am too tired or I happen to notice that there is laundry that really should be taken up to our closet.
We don’t do the traditional Elf-on-a-Shelf, where the kids are told that the red-and-white cloth dolls will watch them and tell Santa Claus who was naughty or nice. No, thank you, Surveillance Elves.
Ours is more like an advent calendar: Every December morning until Christmas, the kids wake up to two little elves who have written them a poem, trapped themselves in the microwave, or done something else to make them laugh.
And so, even though Sima is Supermom, around day 22 or 23, she inevitably calls in the B-team. And the B-team, despite his big talk, occasionally uses AI. And he persuaded his wife to do the same.
So, this year, on Christmas Eve, we decided to use Google’s Gemini and the Elf App to make the elves a farewell video. In the short montage, our floppy fabric elves come to life, make snow angels with powdered sugar stolen from the pantry, and then cheerily inform the kids—while opening a tiny red door in the baseboard—that they were heading home. Sima and I were quite proud of ourselves.
First thing Christmas morning, the kids ran down the stairs, saw the snow angels, and came back screaming and laughing to our bedroom. “Oh wow, guys, I think the elves actually sent you a video,” my wife said, opening her phone to a text from an unknown number.
She clicked the link; the video started, and the kids watched, wide-eyed, as the elves came to life, made their mischief, and then waved goodbye, fading to black as Michael Bublé crooned “Have a Holly, Jolly Christmas.”
My daughter was awestruck. The elves were rag dolls; she thought she was in on the joke. Now … they were real? And they left through that fake-looking door that was Silly-Puttied to the baseboard?
We comforted her: “They went home to their workshop, honey.”
But the spell did not charm my son. He looked at us, looked back at the video, and scrunched his nose like it was a bad smell.
“No they didn’t,” he said. Then he ran downstairs to open his presents.
