Infinite Friction: Why the Nuclear Analogy Is Wrong for AI
Last updated 07:19 AM EDT • Read time: 4m
The conventional wisdom is well-established: the race between the United States and China to build powerful artificial intelligence is most akin to the arms race of the atomic age. Sam Altman of OpenAI famously compared his company’s efforts to The Manhattan Project. Anthropic’s Dario Amodei compared selling advanced chips to China as akin to selling nuclear weapons to North Korea.” Elon Musk went further, saying “…the danger of AI is much greater than the danger of nuclear warheads, by a lot…”
But the nuclear analogy is a limited shorthand reference; it quickly and intuitively situates the gravity of the risk from powerful AI in a framework that people readily grasp. But there is a foundational discrepancy in comparing AI to nuclear weapons; and as the debate over “pacing the frontier” gains traction, the geopolitical shortcomings of the nuclear analogy for AI begin to matter tremendously. Especially in the context of competition with China.
Employment of nuclear weapons are whole integers in strategic equations. Their operative function is discrete: they are either used or not used, crossing the kinetic threshold or not. The decision to employ nuclear weapons was made only twice in human history, in the span of days, bringing a dramatic and frightening conclusion to humanity’s largest war. No other conflict has been able to break the inherent logic of deterrence by mutual destruction, as possible gains from their use are rendered hopelessly pyrrhic in light of the counterstrike.
Employment of AI is fundamentally different; more like a float in strategic equations, existing on an approximation or continuum that makes exact representation limited. It can be used at a high level of totality, seeking a knockout punch. Or it can be used at a low level of severity, running small scale attacks near-continuously, below the escalation horizon.
The Cold War arms race settled into an ironically stable balance of terror because the employment of humanity’s most powerful weapon was threatened but not enacted. In today’s race for powerful AI, it is far more likely to be a vortex of incessant probing, spying, hacking, defending, undermining, stabilizing, calculating, counterbalancing, innovating and deterring. Not a standoff between loaded arsenals that are on alert yet never employed, but rather two colossal systems of systems of intelligence maintaining a million points of friction, operating below the threshold of kinetic destruction.
The speed of such friction is a cat and mouse game, replicated at unimaginable scale and running at a speed that sits an order of magnitude above any prior arms race. Moreover, the advent will not come cleanly, as one nation arrives at the capability before the other. And it is here where the nuclear analogy, serving as intellectual scaffold for its accompanying calls for a mutual slowdown of AI progress will falter. As of yet, there are no strategically compelling off-ramps that can guarantee national security in an arms race built upon infinite friction. The intelligence of powerful AI, being as floats rather than integers in the strategic equation, cannot be meaningfully “capped” in the style of SALT or “reduced” in the style of START. At least not in today’s technology; possibly not for the foreseeable future given the inducements of geopolitical contest.
Therefore, the burgeoning American movement to “pace the frontier” risks following the nuclear analogy to a dangerously misaligned conclusion. Rather than stabilizing discrete capabilities that will be held as a “safe lead” over China’s own efforts, it will kneecap policymaking decisions by tying fluid technological capabilities to brittle and clumsy impressions of Beijing’s own capabilities. For such an effort to be successful requires a degree of transparency and reliability that borders upon naivety.
In the age of powerful AI, a geopolitical plateau of strategic stability will not come in a static balance of power, negotiated in the manner of bygone nuclear doctrines. Stability will come only from the innovative and effective control of the continuum of escalation in powerful AI’s deployment, in both attack and defense, across strategic and tactical points of infinite friction. That effort may prove to be more costly, more difficult, and more unstable than the nuclear arms race ever was.
Garrison Moratto holds a Master of Science International Relations from Liberty University. He is the host of The New Diplomatist podcast and author of the Distant Shores Substack.