One of many current controversies over artificial intelligence centers on “alignment”: A euphemism for “trying to build AI that won’t intentionally kill us all.” That’s important, but it overshadows a more immediate concern: The misalignment not between AI and humans, but between humans. That is, the misalignment of interests between the people who own AI and the rest of us. Even if Sam Altman, the CEO of OpenAI, could figure out a way to make AI his docile servant, his goals would remain fundamentally and irreconcilably misaligned with humanity’s. The right-now policy conclusion is the same for both types of misalignment: We need an immediate halt to all efforts to push the AI frontier further. The next-week policy conclusion, which a focus on apocalyptic disaster obscures (and might be intended to obscure), is that we need iron-strong regulation to make sure AI advances society’s interests rather than merely those of its tech-lord masters.
Only a small handful of men currently dominate AI (they’re all men), and several of the most powerful resemble James Bond villains. Elon Musk, who owns the AI model Grok, stated that his goal is for humanity (or the small portion of it he’d like to preserve) is to abandon Earth for Mars; he’s also the wealthiest person on our current planet. Peter Thiel, co-founder of the defense surveillance firm Palantir, is bizarrely intent on accelerating the End Times. When asked in an interview whether he even wanted humans to survive, he hemmed and hawed for 17 seconds before getting anywhere close to a “yes.” We might be safer in the hands of SkyNet.
The current “good” tech lord is Anthropic CEO Dario Amodei. Anthropic refused to grant the Pentagon permission to use its model, Claude, for universal surveillance and autonomous robotic slaughter. Perhaps Amodei is indeed less dangerous than his rivals. Still, it was barely yesterday that OpenAI’s Altman was himself touted as the most ethical man in industry. Amazon founder and chair Jeff Bezos shares the all-too-common tech lord obsession with abandoning Earth, and has now completed his shift from (in pro wrestling terms) babyface to heel. The nicest thing anyone could ever say about Meta CEO Mark Zuckerberg is that he’s not Elon Musk. And Google, which destroyed its useful search engine to create the infuriating AI model Gemini, long ago jettisoned its founding slogan of “Don’t Be Evil.”
The misalignment between the tech lords and the rest of us is inescapable. Indeed, it’s an example of the fundamental misalignment that underlies our economy: The interests of corporations and oligarchs are at odds with the interests of the people from whom they derive their wealth. To the extent that AI can ever be tamed, it will remain the tool of those with the wealth and power to tame it—and that’s not us.
If AI finally starts turning a profit on its multi-trillion-dollar build-out, where will that money come from? Not from individuals ponying up $20 a month for Claude Pro. It’ll come from corporations—and they won’t be buying it to serve their customers better. They’ll be purchasing it to fire their employees. The replacements will do the formerly human jobs 70 percent as well, for 10 percent of the cost. That will be terrible for the firms’ workers, terrible for their customers, and terrible for society—but a bonanza for their shareholders.
AI will produce some benefits for those of us who aren’t oligarchs, and has (occasionally) already done so. While promised cures for cancer are nowhere on the horizon, AI has greatly improved radiology. There’s a crying need for somebody, something, to revolutionize health care, one of the most expensive and dysfunctional parts of our economy. If AI did this, it could make all of our lives better—but will it? Only if the interests of those who profit from AI are aligned with the interests of those who pay the bills. And they’re not.
UnitedHealth Group, the nation’s largest healthcare insurance company, has already begun using AI—to deny coverage to its customers more efficiently. According to a case currently being litigated, 90 percent of its AI denials are incorrect—and the company knows it. Every healthcare company either already is, or will soon be, replacing human doctors and nurses with AI chatbots. This should surprise nobody: Healthcare companies have already made 15-minute doctor’s visits standard—that is, when your insurance company permits you to see a doctor at all. Far more often, patients are shunted instead to videoconference with a physician’s assistant.
The economics are simple, and ruthless: The less time you spend with a physician, the more money the company saves. The fewer claims your insurer grants, the higher their profits. For-profit corporations provide an ever-growing share of health care, and their managers have a fiduciary duty to maximize shareholder returns. The relationship between a doctor’s bosses and her patients is inevitably misaligned. And it’s the bosses who decide how AI will be used—in healthcare, and in every other sector of our society.
We can easily see a future in which AI is perfectly aligned with its masters, and perfectly misaligned with us. Every Fortune 500 company will consist of just a dozen executives, with AI doing all white-collar work and robots doing all blue-collar work. All “content” will be AI slop: Newspapers will be written by chatbots, overseen by a human editor to press the button that says “publish. Those of us who aren’t executives at OpenAI will live off government handouts; some of us will pick up gig work as data center security guards.
The market can never solve this misalignment between AI’s owners and the rest of us. The only way to solve it is through political action. As long as Americans still have some power over how our country is governed, we need to use it. We can, if we so choose, regulate their industry—as we regulate every other dangerous-but-necessary industry, from nuclear power to high finance to chemical production.
What regulations should we impose to make sure AI is aligned with us? Just like regulations for other industries, these must be hammered out by experts who work for the people, not adopted voluntarily by the companies themselves: We don’t trust the CEO of a company manufacturing hydrofluoric acid to decide where and how to dispose of his toxic waste.
A simple first step would be to immediately halt what the industry calls “recursive self-improvement” in English, that means “AI companies letting their own programs make themselves more powerful.” We’re already far beyond the place where we can easily control AI—why on earth should we keep driving pedal-to-the-metal any further?
Wouldn’t a halt let China surpass us? Highly unlikely: The Chinese Communist Party hasn’t been risk-tolerant of social and political chaos since the Cultural Revolution. Xi Jinping could be persuaded to join a global pause on frontier research—if he believed he could trust America to enforce it.
And could America enforce such a pause? Yes, if we chose to. AI is dependent on real things in the real physical world—particularly semiconductors (which can be interdicted) and data centers (which can be shut down). It would be exceptionally difficult for AI firms to relocate. And if a tech-lord decided to go full-on Bond villain and establish an overseas base to aim for global domination, the solution wouldn’t be a single 007 but a squadron of F-35s.
Does this sound extreme? AI firms admit the existential threat their products pose, with top alignment experts placing the risk of human obliteration within a decade at more than 10 percent. With stakes like these, the greatest threat would be inaction. Far better to fight the battle now, while there are still humans in an F-35 cockpit.
