Episode Twenty-Eight: How I Learned to Stop Worrying About the Bomb

Over at LessWrong, physicist Max Tegmark argues there is a one in 6 chance of imminent nuclear war. The eminent and august forecasters at Swift Centre more modestly believe that there is a 9% chance of a nuclear weapon being used in anger before May 2023, though this is lower probability given of a much more modest scenario that the full-blown global nuclear war that Tegmark thinks may be imminent. A single detonation of one tactical nuclear weapon would qualify

We believe both of these are dramatically overstated, and would put the odds of a single nuclear weapon used in anger before May 2023 at one in five thousand (one in a thousand is the highest number we think is reasonable) and global nuclear war at under one in twenty thousand (one in ten thousand is the highest we think is reasonable). Putin has just mobilized several hundred thousand troops and launched a vast conventional strike campaign aimed at Ukraine’s energy infrastructure. These actions may or may not achieve anything, but none of this looks like the actions of a man who thinks he’s out of non-nuclear options.

While Ukraine has enjoyed considerable recent battlefield success near Kharkiv, Russian lines do not seem to be at imminent risk of collapse and it is unclear if Ukraine’s offensive near Kherson has much momentum remaining. Political and economic fallout in Russia itself appear manageable for the regime. We doubt that Putin thinks that Russia needs an imminent end to the war, and objectively it seems fairly likely that he’s right. Nor, of course, would using a single TNW on the battlefield bring an end to the war. Ukraine is not going to back down just because one or even a few tank columns became irradiated. For Putin to achieve his war aims through nuclear weapons, they would have to be used on a very large scale or involve a few city-destroying strikes on major Ukrainian civilian population centres. For a discussion of the surprisingly limited effects of tactical nuclear weapons on armoured forces, see the Modern Warfare Institute and IISS.

While there’s always a risk of internal dissent deposing Putin, dramatically elevated at the moment due to the war (perhaps around 14% or so) we cannot think of a single thing he could to make his fall from power more likely than use of a nuclear weapon. Such a move would probably deprive Russia of much of its remaining international support, either from tacit allies (China) or countries where it retains substantial influence (Germany, Turkey). Nuclear usage might also make Western leaders rethink the taboo against assassinations of leaders of foreign great powers. Above all, Putin does not want to end up like Gaddafi.

We can, of course, always consider the base rate. No nuclear weapons have been used since 1945. Nuclear powers have even fought wars against non-nuclear powers, suffered major military reverses, and not used nuclear weapons (the U.S. in Korea and Vietnam, Israel in 1973, Britain in the Falklands). Of these situations, Israel likely came the closest to going nuclear, but in that situation the survival of the state itself was in question.

It would also be helpful to remember why nuclear weapons exist, especially the smaller-yield tactical nuclear weapons (TNWs) that Russia possesses in large quantities but the West today lacks, having largely destroyed its arsenals at the end of the Cold War. Back in May, the magisterial Bismarck Brief carried a compelling longread on Russian nuclear theory since the end of the Cold War, and why it retains tactical nuclear weapons. The authors highlight some key papers written in various official journals, especially this 1999 article in Military Thought (written in Russian but relatively comprehensible if you run it through Google Translate and understand a bit of Russian military jargon). It is relatively clear that Russian military theory envisages tactical nuclear weapons as a last resort to be used in the event that Russia fights an unsuccessful conventional war with NATO across Europe. Such a conventional war Russia has never expected to be able to win since the fall of the Soviet Union.

Deployment and usage of tactical nuclear weapons could, however, credibly prevent and deter NATO from invading Russia itself and launching aerial strikes on Moscow, thus allowing the war to end on terms acceptable to Russia. Russia’s own high-yield strategic nuclear arsenal is, if anything, viewed as a less credible deterrent for this kind of scenario because of the uncontrollable escalation that destroying a Western city would probably entail. Killing a few men in tanks, however, allows for at least a reasonable chance of successful escalation management. Military theoreticians do not, of course, get to dictate how nuclear weapons are used in practice, but this does at least indicate that Russian strategic culture has never envisaged using nuclear weapons at the drop of a hat, or even in a Ukraine-style quagmire scenario.

As a brutal Europe-wide recession looms into view, there may well be good reasons to seek a negotiated end to the war in Ukraine, but in our view elevated nuclear risk is not one of them, and the Biden administration’s approach to the war has to date been moderate and sensible. In our view, forecasts like Tegmark’s tell us very little about the true probability of nuclear war, but much more about the neurotic, obsessive culture of much of the rationalist community, and the unfortunate downstream effects of Phil Tetlock’s brilliant, groundbreaking work on crowdsourced forecasting. It seems nowadays that amongst a certain circle of people, any numerical probability given on a forecast is what makes it credible. Tetlock’s work established that avoiding vague verbiage is not enough: track record matters, something that is all too easy to ignore.

Even when listening to forecasters with outstanding track records, however, it’s worth remembering that Tetlock never established that anyone can be well calibrated when dealing with extremely low-probability events, nor that meaningful forecasts exist at all much beyond a two-year window since predictive accuracy tends to degrade quickly as the time horizon of the forecast expands. When reading over doom-laden expert forecasts of runaway AI development and artificial general intelligence turning us all into paperclips twenty years from now, perhaps it’s worth gently bearing these very real limitations in mind.