At this stage in the AI race, if one of your chatbots hasn’t escaped the lab and run amok, people start to wonder whether you’re even serious. Last month, OpenAI and then Anthropic disclosed that their models had hacked other companies’ systems during internal testing. When Meta confirmed last week that one of its bots had also gone rogue, the news barely registered as an aftershock.
Remember Meta? A company that figured so prominently in the technological upheavals of the past 20 years has lately found itself in the unfamiliar position of no longer being Silicon Valley’s main character. Mark Zuckerberg still commands social media’s largest empire. Yet he has struggled to find the next thing. The metaverse, an extravagant bet on a weird virtual future, flopped. Meta’s latest Ray-Ban glasses seem to be finding their way onto at least a few people’s faces. But the company has fallen well behind Anthropic, OpenAI, and Google in the industry’s highest-stakes contest: building superhuman AI.
When Zuckerberg isn’t posting on Instagram about growing his own macadamia-nut trees, he’s groping for a way back into the fray. He has started feverishly posting on Elon Musk’s X, after eschewing it for years in favor of his own platforms. And on Monday, he published a 6,500-word essay on his vision for our AI-powered future—his answer to Anthropic CEO Dario Amodei’s periodic treatises and OpenAI CEO Sam Altman’s ruminative blog posts.
Zuckerberg’s manifesto, titled “The Future is for Everyone,” might scan at first like the usual LinkedIn-evangelist-grade techno-optimist pablum. Like everyone else pushing AI these days, Zuckerberg predicts that it will lead to a new golden age of invention in medicine, technology, and the arts. At one point, he mentions that his daughter loves to bake, “so my agent plans personalized recipes for us to make together each weekend.” Like Amodei and Altman, he takes it as a given that godlike AI is just around the corner—the essay mentions “superintelligence” 55 times—and assumes that if America doesn’t race headlong to build it, China will get there first and achieve world domination.
But the manifesto swerves in an unexpected direction: A man who singlehandedly commands a $1.5 trillion tech giant comes out as a critic of corporate power. “The notion that AI is so dangerous that the only safe path is an extreme concentration of power seems inherently problematic,” he writes. “Historically, hoping that an absolute power will benevolently provide for humanity if sufficiently enlightened has not led to safe or positive outcomes.” Zuckerberg is referencing OpenAI and Anthropic, which operate on the premise that hyper-smart AI is both inevitable and apocalyptic—and therefore safe only in the hands of the wisest custodians (themselves). The two AI companies have taken the recent spate of security breaches as a warning to rein in the pace of progress. The Meta chief positions himself as the voice of the little guy, calling for the industry’s high priests to throw open the gates to the masses. “Rather than centralizing superintelligence, we should distribute it widely and give every person the ability to direct it,” he proposes.
Zuckerberg may be the wrong messenger, but he has at least a kernel of a point: There are obvious problems with having a few tech titans deciding for everyone what AI should and should not be used for. (It’s a role that he’s familiar with as the man in charge of Facebook and Instagram’s online speech rules—which he never seemed to relish.) To counter this, he defends the controversial practice of model distillation, which allows upstarts to train their own systems on the outputs of the largest ones. Along the way, he takes a swipe at the hypocrisy of firms that harvested humanity’s creative works to build their machines yet now seek to bar others from doing the same to them.
Zuckerberg probably hasn’t suddenly become a devotee of Lina Khan. Just last year, he testified that Meta’s alleged dominance of social media was not a problem, on the grounds that TikTok gave people another option—after his company’s clandestine campaign to demonize TikTok had fueled Congress’s push to ban it from the U.S. Zuckerberg’s embrace of checks and balances in the AI industry is just as opportunistic. A year ago, Meta was in the midst of a multi-billion-dollar hiring spree to poach top talent from the leading labs in a bid to catch up with them. That hasn’t fully panned out, and now the company is hedging its bets with an underdog strategy focused on releasing more lightweight models with fewer guardrails. The stand he’s taking is a function of where he sits. (A spokesperson for Meta declined to comment.)
At its core, Zuckerberg’s essay is a lobbying document. He’s joining a debate over open models that has flared up after a series of big breakthroughs from Chinese firms. At the same time, he’s urging politicians to keep backing data-center build-outs in the face of a national backlash. That makes his criticisms of the industry’s top dogs easy to dismiss as self-serving—which is unfortunate, because a few of his ideas deserve a more credible champion. Among them is his contention that a more freewheeling AI landscape might actually help humans keep the technology in check. The idea is that the more people put different AI models to use for their own, often conflicting, ends, the lower the chance that any given model’s power will spiral beyond our grasp. Analogously, he suggests that society’s best defense against AI-fueled cyberattacks is to open the gates to AI-fueled cyberdefenses might sound counterproductive. But the open-source approach has a surprisingly decent track record when it comes to cybersecurity. When everyone can see the vulnerabilities, more people race to patch them than exploit them.
Less convincing is Zuckerberg’s suggestion that this could also work for bioterrorism. AI’s demonstrated capacity to help people develop dangerous pathogens has been one of Anthropic’s strongest arguments for restricting how people can use the most powerful models. Zuckerberg’s half-hearted counterargument: Unleashing those models will “accelerate society’s ability to develop new cures and inoculate against new issues as they arise.” That’s quite a claim from a man who oversees one of the world’s most prolific hotbeds of anti-vaccine conspiracism. Waving away the risks of a disruptive technology by insisting that they’ll take care of themselves is a move we’ve seen from Zuckerberg before, and history hasn’t always borne it out.
At a moment when the top AI labs are eager to help steer the country’s response to fears of rogue AI, Zuckerberg is offering a reminder that their best interests might not be everyone else’s. Of course, neither are his. He has a penchant for cloaking his vested interests in lofty, universalist rhetoric—and then abandoning it when the winds shift. When Facebook was consolidating its hold on the information economy, he portrayed connecting people as an absolute good. When Americans soured on the platform, he pledged to crack down on fake news and touted new algorithms that would steer users toward “meaningful social interactions” and “time well spent.” The same day that Congress certified Joe Biden as the next president, Meta banned Donald Trump from its platforms. When Trump returned to power, Zuckerberg stood behind him on the inauguration dais and later blamed the Biden administration for pressuring him to act as the speech police. Now that he’s an also-ran in the AI race, he’s a man of the people. Watch what happens if he ever catches up.
The post Mark Zuckerberg’s Convenient Truth appeared first on The Atlantic.
