Somewhere in the vast catalog of things people argue about online, there is a corner where the question under discussion is whether nuclear war would kill every last human being, and where the bickering is conducted with the measured calm of people double-checking a dinner bill. The case in point: a detailed investigation published on LessWrong by researcher Jeffrey Ladish, which concluded that even a full-scale nuclear exchange is unlikely — he puts the odds at under 1% — to cause human extinction.
The argument rests on three legs: that the kinetic destruction of blasts and fire cannot physically cover enough ground; that radiation could in theory do the job but in practice would not; and that nuclear winter, while capable of horrific famine, falls short of extinction-level. The thread beneath the post reads like a peer review conducted in public, by strangers, some of them anonymous, several of them clearly dangerous with a napkin calculation.
One reviewer set out specifically to decide whether reading the post justified its conclusion, starting with an epistemic spot check. The claim of 14,000 nuclear warheads in the world? Verified as true. The claim that average warhead yields run below one megaton, probably closer to 100 kilotons? ‘Probably true, possibly misleading.’ The reviewer’s own lookup found the American W78 warhead at 335–350 kilotons, the W87 at 300 or 475 kilotons, Russia’s R-36 missile at 550–750 kilotons and the R29 at 100 or 500 kilotons. So the true safety margin was closer to a factor of three than the factor of ten the original phrasing implied. As the reviewer noted, the great thing about a safety factor is that it’s still a safety factor — but they did grumble about the absence of source links.
Then came the catch of the thread. The kinetic-destruction section leaned on the Earth’s surface area, and a commenter pointed out that the planet’s land area is 148.94 million square kilometers — not the roughly 510 million the post had used, a figure that represents the entire surface, oceans included. Since only about 30% of that surface is land, the post’s area number was 3.3 times too big. The consequence is not trivial: what the post presented as a lower bound on human habitation five times larger than the maximum possible blast area shrinks to just 1.5 times larger, which the commenter called ‘pretty plausible’ wiggle room for worse outcomes. The error was caught, the commenter explained, simply by remembering that the Earth’s circumference is about 40,000 kilometers and working backward. It is hard to decide which fact is more striking: that the apocalypse analysis contained a geometry error, or that it was found by a person doing mental arithmetic at a keyboard.
On the structure of the argument itself, the reviewer was largely persuaded. Kinetic destruction and radiation, they judged, are ‘very solid’ territory — enough weapons tests have been conducted that blast physics are well understood, and fallout dispersal is at least predictable in principle; even if the predicted patterns are wrong, the total amount of radiation doesn’t change, only where it lands. Climate modeling is the soft underbelly. A nuclear winter is far outside anything humans have measured, and as the reviewer put it, different researchers have come to different conclusions, so to trust the conclusion you have to trust the models.
Survival versus everything after
Much of the thread circles a distinction the original post deliberately sets aside: surviving as a species is a considerably lower bar than surviving as a civilization. One commenter argued the deeper problem would be a shattered population concentrated in developing countries suddenly cut off from imports, and raised the oft-debated hypothesis that a post-collapse world could never reboot industrial society because the easily accessible fossil fuels are already gone. Another pushed back, arguing that depletion fears are overblown — a collapsed civilization would inherit mountains of already-extracted raw materials and, more importantly, knowledge — though they granted that Armageddon would be ‘a severe hindrance to technological progress,’ on the grounds that political, social and economic stability appears advantageous to the creation of knowledge. As understatements go, that one has some range on it.
Several commenters appealed to Toby Ord’s The Precipice, which treats ‘knock-on effects’ as central to existential risk: a catastrophe need not kill everyone to count as an existential threat, if it cripples humanity’s ability to handle whatever comes next. One commenter wondered aloud whether survivors, freshly acquainted with the fragility of life, might emerge more committed to safeguarding the future. They answered their own musing: no. We know the dinosaurs went extinct, the average mammalian species lasts about a million years, and life is precious and unreasonably rare in the universe — and yet, in aggregate, existential risk reduction doesn’t factor heavily into how we allocate our resources.
There’s also the uncomfortable observation that nuclear war would probably not arrive alone. One commenter noted that countries trading all-out nuclear strikes are likely already in a world war, possibly with chemical or biological weapons in the mix. Ladish replied that he expects conventional, chemical and biological effects would be dwarfed by nuclear ones, and that deterrence logic works against secret WMD programs — ‘the whole point of a deterrent is lost if you keep it a secret.’ Another commenter offered the bleakest comfort in the thread: biological and nuclear warfare don’t mix well, because without commercial air travel and trade, biological agents don’t spread well.
The thread never arrives at a verdict, because threads like this never do. What it offers instead is a rare spectacle: the end of the world, treated as a solvable word problem, audited line by line — and the single most consequential correction delivered by the humble fact that the Earth is about 40,000 kilometers around. In the geography of the apocalypse, seven in ten squares are ocean, and nobody lives there.