AI and Nuclear Proliferation

As long predicted, the world order is shifting, away from American unipolar dominance and toward a system of increasingly bold middle powers and a new superpower and true U.S. peer in China. We are observing a diffusion of power in the international system and, with that, an emerging multipolar order characterized by the decline of U.S. global hegemony, increasing geopolitical complexity, new military capacities, and growing economic and financial partnerships (for example, the BRICS group). There can be little doubt that within the coming years, the nuclear club will add new members. More than ever, the lovers of peace and humanity need to be prepared to press the case for nuclear disarmament and robust arms control treaties that seriously limit nuclear arsenals and forestall the development of new nuclear weapons.

An increase in the number of countries with deliverable nuclear bombs significantly raises the odds of an existential-level disaster involving a nuclear exchange. Recent analyses warn that several key factors have come together to increase the risk of both intentional and accidental exchanges, as well as miscalculations. Ultimately, nuclear proliferation (together with other surrounding factors like updated arsenals and shorter missile flight times, high-powered cyber attacks on networks and energy supplies, and new geopolitical variables) dramatically increases the risk of intentional use, miscalculation, and an accidental nuclear exchange. Donald Trump does not understand that attacking people with punishing violence, killing thousands of innocent people and destroying their country does not make them want to negotiate, whatever that could possibly mean now. No one, apparently, taught Donald Trump that this is not a way to negotiate. He may believe that he is in full control of the situation, but many global experts believe that his actions (and Benjamin Netanyahu’s) have increased the likelihood of a nuclear attack.

Like generations of presidents before him, Donald Trump believes that it is the responsibility of the United States to shepherd world affairs through shows of military strength, if only to protect its position as the world’s most powerful empire. Some of the reputedly smartest and most polite in the world believe this, though they disagree with Donald Trump’s style of management: his many and frequent disrespects and harsh words for some of Washington’s oldest and strongest allies; his abuses of the global economy as a mere pressure tactic, as if this makes other countries want to world with us more; his government’s many wars of choice, under whatever the insulting euphemism (operation, excursion, regime change, military action, whatever); his contempt for both our constitutional traditions and international law; his domestic repression and authoritarianism; and his hatred of our immigrant friends and neighbors, among others.

Questions about the proliferation of nuclear weapons are also deeply connected to the speculative jubilee around AI. As has been widely reported, the energy consumption associated with running Big AI’s systems has already placed acute strains on the electrical grid. Given the widespread sense that all of society should accommodate Big AI at breakneck speed and with reckless abandon, the question has turned to how to get the most energy as fast as possible. Unsurprisingly, this enthusiasm has meant an increase in interest in nuclear energy to propel the rapid expansion of AI’s real-world infrastructure. As the Department of Energy notes, large-scale AI systems require continuous and high-load electricity for training and inference, and large data centers have started to strategically prioritize firm power sources.

To many, nuclear energy fits the bill in that it gives you stable load generation with low direct carbon emissions. Thus many tech firms, along with their policymaking tools, have demonstrated their renewed interest in extending the life of existing nuclear reactors and fast-tracking the development of the newer designs. Last year, the White House announced the Genesis Mission as “as a dedicated, coordinated national effort to unleash a new age of AI‑accelerated innovation and discovery that can solve the most challenging problems of this century.” A recent Department of Energy report on the Genesis Mission sheds some light on the connection between the latest tech boom in AI and reinvigorated calls for nuclear power:

The DOE report noted that nuclear power plants have historically been challenged by long development timelines and high costs, while demand continues to grow, including from AI data centers.

To meet this challenge, the DOE proposed the use of AI to design, license, manufacture, construct, and operate nuclear reactors “with human-in-the-loop workflows, enabling at least 2x schedule acceleration and greater than 50 percent operational cost reductions.” These goals are being addressed by the DOE through research into such AI technologies as surrogate models, agentic workflows, autonomous labs, and digitals twins.

The speculative fervor around AI offers the perfect cover for many countries to fast-track nuclear programs. They can claim that they are only trying to compete in and partake of the so-called AI revolution, as they build up the enrichment and reprocessing infrastructure necessary for a usable weapon. It may be that we will also soon count nuclear proliferation as a consequence of AI mania. We will have more nuclear countries, more variables, more unknowns. There is an inherent connection between developments in the direction of nuclear power and weapons. All governments understand this and acknowledge it, even if ordinary people and supposed experts frequently do not. As Professor M.V. Mavana recently wrote, discussing the remarks of Saudi Arabia’s Mohammed bin Salman, “Why are these statements significant? Because there is a long track record of attempts by the nuclear industry and advocates for nuclear power to erase or at least camouflage the connection between the technologies used to develop nuclear energy and the capacity to build nuclear weapons.”

The conditions for state actors in the real world have been such that they must ignore our delicate distinctions. Some experts believe that a “limited nuclear exchange” would kill billions of people. But the frightening truth is that they don’t really know. Scientists cannot model any of this well enough to know. And to this point, the record of predictions has been very poor. Nuclear scientists have consistently failed to predict the size, the destructive capacity, and the long-term health and safety consequences of nuclear explosions. The Castle Bravo test was the first in a series of extremely destructive nuclear tests, and it makes a good example of a bad “expert” prediction:

The reason for the unexpectedly high yield was due to the “tritium bonus” provided by the lithium-7 isotope which made up most of the lithium. This isotope was expected to be essentially inert, but in fact it had a substantial reaction cross section with the high energy neutrons produced by tritium-deuterium fusion. When one of these high energy neutrons collided with a lithium-7 atom, it could fragement [sic] it into a tritium and a helium atom. Tritium was the most valuable fusion fuel, being both highly reactive and causing extremely energetic fusion, so this extra source of tritium greatly increased the weapon yield.

The scientific experts had modeled out a linear fuel depletion scenario, but they touched off an exponential fuel accretion scenario they did not expect. They were children playing with fire. There is no doubt that practical nuclear science has come a long way since the mid-1950s. But the number of potentially catastrophic unknown variables has also only increased. The oligarchs are richly incentivized to build and use the weapon, and at such heights, the science community and the military and political elite are effectively one and the same. There is a problem built into the structure of the system. This is why every country without these weapons hastens to acquire them, hoping for any bargaining posture. Pretensions aside, all countries that want a nuclear arsenal want it for the same fundamental reason; they see such weapons as the key to their future. The weapons that could destroy humankind are seen as necessary to the strategic plans of the world’s most powerful countries. Now, that’s really something to consider.

But perhaps a still more under-appreciated and troubling aspect of this situation is the epistemic asymmetry that separates us, the popular masses who will carry the costs and bear the consequences, from those pushing for things like AI data centers and increased nuclear capacity. The ruling classes who design nuclear weapons, build nuclear reactors, and develop the AI systems now accelerating both do not have proportionate exposure to the downside risks of their choices. This asymmetry of power is core and constitutive to the system’s behavior. The structure of incentives explains the system and its outcomes. The political and economic system is organized to externalize catastrophic risks onto hundreds of millions of people who effectively have no political representation and no real voice in policy or governance. We are watching the compounding of this externalization of costs to an unprecedented degree, and at a time of rapid change and uncertain technological capacity.

The argument is a modest one: in cases where decision-makers are insulated or removed from the worst-case outcomes of their choices, the political-economic system will systematically underprice the catastrophic tail risks. And if this asymmetry is constitutive, and not incidental, then these bleak results do not arise from bad actors. They come from the architecture of how decisions get made by a ruling class. In the world we have come into, the opponents of war and nuclear weapons are more needed than ever, and they must understand powers aligning in Big Tech, the arms industry, and sadly much of the science community.

The post AI and Nuclear Proliferation appeared first on CounterPunch.org.