The narrative pushed forwards by top US AI labs, and now passionately embraced by the Trump administration, is that America must win the AI race at all costs. Being first to develop Artificial Superintelligence (ASI), the argument goes, would usher in a new era of unparalleled American global dominance across practically all sectors—while losing the race would be tantamount to handing total control of the world order to Beijing. The government is therefore adding its own weight to AI companies’ existing hundreds of billions of dollars in annual AI spending, with Trump publicly announcing that “America is the country that started the AI race. And as president of the United States, I'm here today to declare that America is going to win it.” At the same time as winning the AI race is becoming a top national priority, experts from around the world are near-unanimously raising the alarm that this policy of unfettered accelerationism will likely bring about the imminent extinction of the human race. However, this overwhelming scientific consensus has been shrugged off by much of Washington, likely due to its lack of intuitive appeal to a non-technical audience. In this essay, I therefore choose to set misuse and misalignment concerns aside in favor of a more intuitive geopolitical analysis. This will lead to the conclusion that even if AI safety experts were wrong, and even if American ASI were guaranteed to end up contained and docile and in the right hands, pushing to win the race is nevertheless a major blunder by our government all but guaranteed to lead to catastrophe for the American people.
——
Imagine a hypothetical scenario taking place several years from now, but with the tables turned. Despite American researchers’ best efforts, China is ahead in the AI race. How far ahead exactly is painfully unclear: while the latest publicly released models seem roughly six months beyond the US frontier, the few American informants still remaining in the increasingly-secretive Chinese AI industry report rumors of far superior models nearing the end of development. Meanwhile, on the international stage, Beijing refuses to engage on the topic of AI governance, only further reinforcing suspicions. It begins to dawn on Congress that a Chinese ASI could realistically arrive at any moment—and that where a conflict with China might be winnable today, a war against a superhuman intelligence wielding the nation’s vast human and capital resources is bound to play out as one of the most one-sided in human history. Although it is not clear exactly when America becomes captive to the whims of Beijing if things are allowed to proceed, this outcome seems no less guaranteed.
What is to be done? Pushing to catch up is unlikely to have any effect: not only has the brunt of America’s resources already been brought to bear in the past few years, but even a successful reduction of China's lead down to several months would hardly dampen the impact of its eventual victory. Timelines involving superintelligences thinking thousands of times faster than human experts are simply far too short. Convincing the enemy nation to abandon its hard-earned path to global dominance appears equally impossible, especially with a geopolitical rivalry aggravated by years of brutal competition. It seems that only one option remains to ensure US security: an ultimatum demanding a moratorium on Chinese AI development…and failing rapid compliance, a preemptive nuclear strike. Drastic as it certainly is, this suggestion initially draws humanitarian opposition. But filled with some of America’s most outspoken patriots, many growing up against the backdrop of the Red Scare and building careers off of American exceptionalism, Congress slowly comes to a somber agreement: better risk nuclear war than let the nation fall into the grips of totalitarian AI.
——
In the real world, we are likely still several years away from this scenario. Even so, we seem poised to go down an exactly identical path.
- America currently holds the lead over China in the AI race, and is preparing to do whatever it takes to maintain it. This not only takes the form of spending untold billions of dollars and throwing environmental concerns to the winds, but also vastly enhancing secrecy and information security within the industry while taking the first steps to prepare for an AI-powered military.
- The Chinese government’s own commitments to the AI race reveal that Beijing is alert to the unparalleled power that comes with controlling a superintelligence; moreover, the Sino-Russian bloc is perhaps even more passionate than the US when it comes to putting sovereignty and national security above all else. Adding in the effect of constantly mounting international tensions, American-controlled ASI would certainly be considered an existential threat to be avoided at any cost.
- While peaceful negotiations would be the obvious first try, the Trump administration has already made abundantly clear its unwillingness to come to the table with the CCP. Just several days ago, for example, the White House’s sweeping AI Action Plan admonished international efforts to impose “burdensome regulations,” and instead recommended that the government “vigorously advocate for international AI governance approaches that promote innovation, reflect American values, and counter authoritarian [Chinese] influence.”
- With this option off the table, it truly seems like an endgame is approaching where China and its allies are left with no choice but to meet continuing US ASI research with the only sufficiently punitive tool at their disposal: nuclear weapons.
To some, desensitized to nuclear posturing by the Cold War, it may seem as though Beijing would never follow through with such a plan. After all, the US went through decades of competition with the USSR without triggering nuclear war—even when the American weapons program pulled well ahead of that of the Soviets. However, deterrence only ever held in the Cold War because there were no sufficiently capable nuclear defense technologies: no matter how big of a bomb the Americans might have threatened to deploy, a counterattack by even a relatively underdeveloped Soviet nuclear program would still have been more than enough to level every major American population center. In other words, regardless of how much it fell behind, the USSR was untouchable. But now imagine if the US had started construction on a device that could swat all Soviet nukes out of the sky, thus nullifying the USSR’s only guarantee of safety. Clearly, this would have forced the Soviets into drastic action before they forfeited their only source of deterrence! This is exactly the case with ASI. Whether through scientific innovation, advanced cyberwarfare, or even methods currently unfathomable to humanity, an American superintelligence threatens to cripple China’s counterattacking capability and strip Beijing of its sovereignty, demanding even the most extreme preventative measures by the CCP before this becomes a reality. American victory in the AI race does not guarantee a world of plenty, spreading on the wings of global democracy—it guarantees our very nation’s own nuclear destruction.
——
There is one way to prevent this scenario from materializing: a halt to the race, in the form of a global moratorium on AI development. Even with this attempt to reduce catastrophic risks, it is true that the threat of eventual nuclear repercussions would likely still prove necessary to ensure compliance. However, this scenario of an early moratorium is nevertheless far superior to the one predicted by the status quo in that it gives the world a cushion of several years to peacefully coordinate for the future.
Having this time is critical for the creation of a powerful international governing agency to monitor AI research and ensure full global compliance. We are indescribably fortunate that AI models require massive computational resources to train, putting the moratorium’s enforcement well within the realm of possibility, but this remains a task measured in years, not days. Having this time also enables close cooperation with other nations from around the world, legitimizing and future-proofing the moratorium by promoting it from a US-China treaty to a global accord; this also strengthens political pressure on countries reluctant to comply, reducing the risk of nuclear reprisals becoming necessary. Having this time serves as well to make sure that nuclear weapons need not be deployed in the early years of the ban, leaving time for violation reverification and negotiations while reducing the risk of a false alarm. After all, if there is one lesson to be learned about managing nuclear deterrence from the Cold War, it is that even systems designed for nuclear-grade robustness are often all too fallible. A safe way out of this crisis exists, all we need is time—so the sooner the existential risks of AI development are taken seriously on the world stage, the more likely humanity is to endure.
Fortunately, being ahead in AI puts America into the optimal position to kickstart this process by leading with proactive, credible commitments towards the moratorium. Not only would this eliminate the threat of preventative strikes directed at the US by proving that American ASI is not in development, but by the very same token, it would immediately vanish China’s own existential incentive to accelerate AI development. Forced to choose between a peaceful moratorium and nuclear war, the nation would certainly join the US in a détente.
——
In reality, leading the charge to pause ASI development is not a matter of the United States making sacrifices to benefit the rest of the world. Rather, it is American lives, America’s own future that hang in the balance. Superintelligence can wait: the technology will never disappear from within our reach. So instead of the current policy of blindly sprinting towards the finish line, our government must first do everything possible to ensure that an America will even exist to reap the rewards of a victory. Not becoming the primary target of the world’s first nuclear war comes first; and only then—when international trust reemerges, when this can take place openly and safely, when victory no longer means certain death—can the governments of the world work together to unlock the extreme benefits harnessing superintelligence will bring, both for America and to the entire human race.
