The author rejects the idea that advanced AI could “put the world back together” after a nuclear war. A Nature Food soot-injection model estimated about a 7% drop in global crop calories in a 5-teragram case and roughly 90% by years three–four in a 150-teragram case, with possible famine deaths in the billions—figures the piece treats as model risk, not prophecy (National Academies 2025).

AI’s best nuclear role is before launch: widen the picture, surface contradictions and alternative explanations, and shrink authority when sensors disagree—not compress time to fire. After a blast it could map smoke, food, ports, and power, but it cannot open a border or rebuild a bridge.

Alternative foods (e.g., an Earth’s Future seaweed study suggesting up to ~45% of demand in 9–14 months if farms can be built) and small, offline systems matter because compute dies with the grid. Restore is the wrong word: AI can help see what remains; it cannot remake sunlight or topsoil. The world worth saving is the one never burned.

Artificial intelligence could help slow escalation and preserve food systems after catastrophe. It cannot replace sunlight, infrastructure, or human judgment, which makes prevention its highest-value role in nuclear risk.

The most dangerous fantasy about artificial intelligence may be less dramatic than the familiar fear that a machine will destroy the world. It is the belief that, after humans destroy enough of the world ourselves, sufficiently advanced AI will be able to put it back together.

A nuclear exchange could end quickly while its environmental and agricultural consequences persist for years. Large urban fires could inject soot into the upper atmosphere, reducing sunlight, cooling the surface, altering rainfall and disrupting food production far from the states that fought the war.

A widely cited Nature Food study modeled six soot-injection scenarios. In the 5-teragram case, the decline in global crop calories was about 7 percent, a shock larger than the biggest anomaly in the FAO production record since 1961. In the 150-teragram scenario, global crop-calorie production fell by roughly 90 percent three to four years after the war. The authors estimated that famine-related deaths could reach billions in the larger scenarios.

Those figures are model-based estimates rather than forecasts. Fire behavior, weapon employment, target selection, weather, smoke injection and many other variables remain uncertain. But uncertainty does not erase the risk. The National Academies’ 2025 assessment examined consequences unfolding from weeks to decades and emphasized how many links in the causal chain still require better data and model comparison.

Use AI to slow the irreversible decision

The highest-value role for........

© Eurasia Review