Summary

  • I find there have been ten accidental nuclear incidents in history where I think there was some non-negligble chance that, had things played out differently, there could've been a nuclear exchange.

  • I think the chance of another accidental nuclear incident happening in the next decade (2022-2032) is ~28%. I take a ~5% chance that any given incident will escalate into a nuclear exchange causing at least one fatality[1], putting the chance of an accidental nuclear exchange in the next decade at ~2%.

  • By combining my estimates here with other averages of expert forecasts, I think the overall chance of nuclear exchange (intentional or unintentional) this decade is ~6%.

  • The 1960s represented a high point in the risk of accidental nuclear exchange. The number of incidents has gone down significantly since 1960 and has gone down significantly again after the 1990s, with no such incidents occuring since 1995.

  • Better technology to reduce false positives has likely reduced the risk of accidental nuclear exchange. Many past incidents were the result of computer errors detecting weather or other peaceful events as nuclear launches and these seem less likely to happen now.

  • More time spent between the development of offensive nuclear weapons and now without observing an incident should gradually over time reduce our chance that such an incident will occur in the future, all else being equal. This makes the risk higher in the past than the future, all else being equal.

  • Thanks to international arms control agreements, we have moved from a height of over 63,000 nuclear weapons in the 1980s to under 14,000 nuclear weapons today. This also, all else being equal, reduces risk and shows that progress is possible.

  • We also have much more geopolitical peace since the end of the Cold War and this has likely reduced the chance of accidental nuclear war. But it's possible this era of peace is backsliding given recent tensions with Russia (and also China).

This is a link post for my new blog, so read the rest there!

66

6 comments, sorted by Click to highlight new comments since: Today at 7:25 AM
New Comment

I disagree with the claim that the overall accident risk is going down. While it's probably true early warning systems are getting more reliable (though the actual degree of this is really hard to gauge due to their complexity)[1], a third party (China) adopting launch on warning arguably raises the risk at least 50%, if not more due to initial kinks. Also, as many have pointed out, the emerging trilateral dynamic of three nuclear peers is unprecedented in history and less stable.

Also, what would count as an accidental nuclear war? I think e.g. the US launching a large salvo of low-observable cruise missiles deep into the Chinese mainland during a conventional war could easily be mistaken as an attack on the silo fields and trigger a nuclear launch.


  1. What I mean by this is it's not like EW systems have been static and only the sensors have been refined over time to make them more reliable, they have been made into ever larger informationized networks etc. and it's not at all clear that the risk of a false alarm generated by any one part of the system is significantly lower now. For examples on how these more complex systems have more points of failure see e.g. this ↩︎

Thank you very much for writing this. I broadly agree with your post, but I probably put less weight than you do on the historical record. I think the crux is that I assign a higher probability to there being close calls that we don’t yet know about, but which would make the picture look very different.[1] Here are a few reasons for thinking this:

  • The post says that documents generally get declassified after 25 years, implying that we would hear about them eventually. This is a US rule though – other countries are less transparent, suggesting that we are undercounting their close calls.[2]
  • Even if documents do get declassified, it’s non-trivial to actually identify close calls. In “The Limits of Safety”, Scott Sagan documents several nuclear close calls.[3] He says that the cases he discovered were probably “just the tip of the iceberg” (p270) – essentially because the archive work to find them was so difficult.[4]
  • It wouldn’t necessarily be obvious to researchers that a given incident was a close call. Nuclear weapons are often thought about in terms of normal accident theory, which posits that complex systems fail in unpredictable ways.
  • Out of 10 close calls, 3 relate primarily to the USSR or Russia, 1 relates to both the USSR and the USA, and  6 relate primarily to the USA.[5] I can imagine that the USA and USSR/Russia would have a disproportionate share of close calls (for example because they have by far the most nuclear weapons). But it would be very surprising to me for the US to have 6 such cases, while the other nuclear weapons states – China, France, Israel, India, North Korea, Pakistan, and the UK – combined  have 0.
  1. ^

    This depends upon the claim that the cases that we  know about are not necessarily representative of the entire universe of cases. 

  2. ^

    A similar point is made in this report from Chatham House (p3).

  3. ^

    Sagan is interested in a broader range of close calls than the one in the post, but from memory I think Sagan was the first researcher to publicly identify the “Missiles over Georgia” and “power outage” cases.

  4. ^

    Sagan also did interviews with relevant people, and submitted requests under the Freedom of Information Act. These are described in the introduction to Limits of Safety.

  5. ^

    USSR/Russia: Petrov, Able Archer, Norwegian Rocket. USSR and USA: Cuban Missile Crisis. USA: remaining 6.

How do you think the push to replace humans with AI systems in nuclear warfare decision making will affect the chance of accidental nuclear war going forward? I hear some countries have been considering it.

What is the benefit of using an AI here?

> Will a nuclear incident lead to a nuclear exchange?
> If we take rule of succession with a 10% naive prior and update based on surviving ten accidents so far, the chance of the next incident resulting in a nuclear exchange is (1-(10+10)/(10+11)) = 5%.

I'm not sure you can do this, because of anthropic effects (you don't get to observe some words with nuclear exchanges). I would adjust this upwards.

Also, a 10% naïve prior smells a bit high, personally.

“Seven of these events involve computer errors mistaking innocent things for incoming nuclear weapons and three of which involve misinterpreting enemy actions as signaling potential nuclear intention.”

Making safer computer systems for nuclear missile detection and deployment seems like a potentially impactful career goal. I know next to nothing about the topic, but some amateur thoughts here: https://forum.effectivealtruism.org/posts/7ZZpWPq5iqkLMmt25/aogara-s-shortform?commentId=rnM3FAHtBpymBsdT7

Great post, good luck with the new blog!