America couldn’t govern AI if it wanted to

The protests against data centers and Flock cameras are not about data centers or Flock cameras. They are about dread over artificial intelligence. (As I type these words at a coffee shop, the man at the next table is talking about how his son uses AI to decide what kind of pizza he wants.)Some chalk the negative vibes up to “doomerism” — and with justification. The Washington Examiner’s David Harsanyi recently wrote that big technological leaps nearly always come with a panic, and the darkest prophecies rarely pan out. What’s more, as Harsanyi notes, the loudest alarms seem always to come from AI executives themselves who sound them to exaggerate their own importance, generate interest, and undermine competition.That’s all likely true. But so are the risks, as Harsanyi later admits. Playing with fire Among the more legitimate ones is AI’s potential to lower the expertise barrier for building bioweapons and launching cyberattacks on critical infrastructure. The more extreme scenarios where AI systems slip beyond human control and pursue their own goals can’t be ruled out, but they aren’t the most pressing worry. The far more likely catastrophe is evil humans misusing AI to carry out evil human deeds.There’s also real concern it could accidentally trigger a war. This spring, a chatbot’s faulty read of a Chinese ship’s cargo sent American military aircraft scrambling and nearly led U.S. troops to board the vessel.We’re playing with real fire here.And these are only the immediate dangers. The longer-term dangers of the AI revolution range from mass unemployment to mass surveillance to the possibility that we dramatically weaken our own cognitive abilities. Twenty-five years ago, most adults could read a map and remember a route; today, many are helpless without GPS. AI threatens to do the same to nearly every mental task we give it, from writing to arithmetic to basic decision-making and reasoning.I’m not above any of this; last Sunday before church, I used AI to pick out a tie. How many repetitions would render me unable to match colors?To Harsanyi’s point, industry leaders and politicians calling for regulation almost certainly have ulterior and self-serving motives. Forgive me if I don’t treat Bill Gates, who spent years cheerleading the technology, as a disinterested prophet of AI doom. Or if I don’t take the Silicon Valley weirdos and cultists at their word.But that doesn’t mean they’re fundamentally wrong about the catastrophic dangers we face today. They may simply know that AI can no longer be reined in, which makes calling for regulation a costless way to appear responsible — and a convenient alibi for when the first real catastrophe hits.Many across the political spectrum have panned the Trump administration as reckless for its no-brakes policy in AI development. I’m not one of them, despite believing in the dangers of AI. I just don’t think there’s a better option.The China problem Before we even get to the catastrophic risks, there’s the China problem. The Chinese Communist Party is one of the most malign political organizations in human history, guilty of genocide against its Uyghur Muslim population, mass surveillance and censorship, the suppression of political opposition and religious freedom, debt-trap diplomacy across the developing world through its sham Belt and Road Initiative, and far more.Most estimates place China somewhere between four and 12 months behind the United States on AI development. Should China overtake the U.S. in AI supremacy, it would be a history-altering event. It cannot be allowed to happen. Under any circumstances. Even if it means allowing AI to be developed in an irresponsible manner.(Getty Images) We must take the chance, however remote, that the free world can develop AI in a way that avoids major catastrophes and delivers great benefits to humanity. Bioethicist Charles Camosy, whom I interviewed on the topic this week, believes that if we “get past the existential and catastrophic risk part of this,” a new golden age of sorts awaits us on the other side.But if China beats the U.S. in the race to superintelligence, there will be no happy ending. We know what the CCP would do with that power because we’ve seen what it has already done. It would export its surveillance state to the rest of the world — and powered by AI, it would have the means to enforce it.Some, including Camosy, are hopeful that the Catholic Church can step into the breach and leverage its status as a moral voice for 1.4 billion people to broker a truce between the superpowers and negotiate a treaty. As a Catholic, I won’t mock this, despite the Church’s failed attempts to reason with the CCP in recent decades. China has its own official, state-sanctioned Catholic Church, and many of its millions of Catholics worship the true Church in secret, underground communities.Neither will I hold my breath for these diplomatic efforts to bear fruit.But suppose I’m wrong about Beijing. Max Tegmark of MIT, for instance, argues that China has no more interest than we do in building something it can’t control, which leaves room for a deal. Even then, the idea that America could govern AI in any meaningful sense betrays a misunderstanding of what regulation requires and how the technology works.AI’s moral formation Regulating AI isn’t a matter of programming it to perform a specific function — the way we program our alarm clocks to go off at a certain hour in the morning. It’s a matter of training it to make moral choices, to choose between good and bad in order to ensure AI’s “alignment” with human goals.To be sure, Washington can test models for bioweapon and cyberattack capabilities, restrict chip exports, and require companies to report incidents. These are worth doing.But those measures don’t touch the problem at the heart of the regulation debate. For now, we can help stop a model from building a pathogen. But we can’t yet teach it right from wrong. And as these systems grow more capable than the people overseeing them, outside restrictions will only work as long as the AI itself has no desire to get around them. What’s really needed is moral formation.Chris Olah, who leads the Anthropic team responsible for understanding how AI systems like Claude act, talks about training AI the same way parents talk about raising children.“How do you help them be stable? How do you help them to mature? How do you help them be, you know, deeply moral?” Olah was recently quoted as saying in the New York Times.Those are a parent’s questions, not a computer programmer’s. And as any parent could tell you, answering them is no simple task. It requires persistence, patience, and perhaps most of all, clarity. You can’t train someone in a specific moral program if you don’t understand it yourself. And America at present does not possess anything resembling a clear moral framework.Anthropic’s in-house philosopher, Amanda Askell, told the New York Times that she wants the models to be “the best of us,” and to be able to discern what is in humanity’s best interest.It’s a noble aim, but with a fatal flaw in its conception: Humanity itself lacks a consensus on what’s in its best interest. It offers many different interpretations on how to answer that question, and many, if not most of them, are mutually exclusive. In America alone, for instance, we can’t agree on when life begins, what marriage is, whether a man can become a woman, whether God exists, or on nearly any other important moral question.So whose version of morality do we expect to instill in AI? And how could we possibly write it into law and enforce it?I put this question to America First Policy Institute’s Joel Thayer in a debate I moderated between him and Juan Londono of the Cato Institute earlier this year. Like every other smart person I’ve posed the question to since, he didn’t even attempt to answer the practical questions. Like others on the Right, he merely repeated his desire that we program AI according to Christian and patriotic moral norms.A recent Phillip Lamb essay in the American Mind does much the same. It correctly claims that there is no such thing as a morally neutral AI model, and that AI should be programmed according to the Western moral inheritance. True enough! But if America itself has no consensus on whether that inheritance is worth keeping, how can we expect to pass it on to a machine?Is it really plausible that Washington can nudge AI toward Christian and patriotic morality without massive political blowback? Or that the next Democratic administration won’t use the same levers to nudge it the other way? Of course not. It sounds nice, but on a practical level, it’s a silly suggestion.I understand, however, why it’s the answer people like Thayer and Lamb give: It’s the best one we’ve got, and it’s utterly unrealistic.IS AI CONSCIOUS?The only realistic path forward is the one President Donald Trump has taken: Unshackle Silicon Valley, beat China, and hope we survive long enough to reap the benefits of AI.My guess is that sooner or later, there will be a catastrophe that scrambles all of these calculations and forces humanity to make hard choices we can’t yet foresee. I, too, wish that safely developing AI were a simple matter of passing a law and enforcing it. But it appears the best we can do is pray the catastrophes are small — and that we beat China along the way.