In your talk at EAG you said that you think the risk of nuclear war today is “high and rising”. You also estimate the annual probability of a catastrophic nuclear event is about 0.5%. I wanted to first say kudos for quantifying your beliefs in this way. It’s so helpful for communicating clearly about these risks. I have two related questions:
(1) Could you please say more about the main considerations, metrics, and/or data you use to inform this estimate?
(2) How quickly do you think the risk is rising? I’m curious whether you think the annual risk is likely to increase by some tenths of a percentage point, or by factors of 2 or more.
Thanks for this question Stephen. Let me start with my overall perspective on the probability of use: Over the ‘deep time’ periods about which I have learned much from the EA Community—periods of hundreds or thousands of years – I believe a CNE is a near certainty unless we dramatically reduce this risk. For a while, states acquired nuclear weapons roughly at the rate of one every five years; this rate slowed after the Nuclear Non-Proliferation Treaty was negotiated, but has never stopped. Over the next century, we will either manage the risks of nuclear technology more effectively, or the number of actors with nuclear weapons will increase until someone makes a mistake. After that, our ability to manage these risks may be destroyed completely.
That said, I also want to share some more insight into the 0.5% figure that I shared at the EA Global meeting. I thought it was important for me to offer the EA community my own assessment about the probability of a catastrophic nuclear event (CNE).
I want to be clear, however, that any probability assessment in this space is, of necessity, a judgment call, no matter who provides it. The nuclear “system” is a massive, complex, system of many sub-systems, and it’s impossible to quantify the risks of use the way one might, for example, conduct a probabilistic risk assessment of the likelihood of failure of a specific engineered technology (where specific data about system performance over time can be employed to calculate probabilities). With the nuclear weapons system, we are talking about thousands of individual weapon and delivery systems, tens of thousands of humans in the decision-making and launch execution loop, hundreds of thousands of technical (and digital) components, and constantly evolving political dynamics and uncertainties which are not quantifiable beyond a qualitative judgment.
Ultimately, I’ve given you my judgment about the probability of nuclear use, on average, in any given year based on:
my experience working within the nuclear system and my understanding of the system design and its vast complexity;
my understanding and experience of the culture of control of the weapons and its weak points, (for example after the cold war ended, the culture of nuclear control in the US military declined significantly – see a report on this here: https://www.hsdl.org/?abstract&did=486958)
my understanding and experience of global politics and the dynamics of great power conflict.
The key point I wanted to illustrate with my probability assessment is that even if we think the risk of use is small in any given year, the cumulative risk is unacceptably large over time. I did not say, but perhaps should have highlighted, that nuclear risk is not static, and some years my judgment about the risk would be much higher than 0.5%. For example, we are watching Russian forces mass along the border with Ukraine as I write, and tensions are very high right now between the US/Europe and Russia. In this moment of heightened risk, and until we back away from posturing our forces for conflict, I would say the risk of use is much higher, perhaps by as much as a factor of 10. If that persists over time, and between more nuclear adversaries, the probability of use over time becomes much higher. Sadly, I suspect that’s the world we are heading for – an increasing use probability – if we don’t succeed in dramatically reducing risks through a variety of measures that are policy, technical and political in nature.
Thank you for doing this!
In your talk at EAG you said that you think the risk of nuclear war today is “high and rising”. You also estimate the annual probability of a catastrophic nuclear event is about 0.5%. I wanted to first say kudos for quantifying your beliefs in this way. It’s so helpful for communicating clearly about these risks. I have two related questions:
(1) Could you please say more about the main considerations, metrics, and/or data you use to inform this estimate?
(2) How quickly do you think the risk is rising? I’m curious whether you think the annual risk is likely to increase by some tenths of a percentage point, or by factors of 2 or more.
Thanks for this question Stephen. Let me start with my overall perspective on the probability of use: Over the ‘deep time’ periods about which I have learned much from the EA Community—periods of hundreds or thousands of years – I believe a CNE is a near certainty unless we dramatically reduce this risk. For a while, states acquired nuclear weapons roughly at the rate of one every five years; this rate slowed after the Nuclear Non-Proliferation Treaty was negotiated, but has never stopped. Over the next century, we will either manage the risks of nuclear technology more effectively, or the number of actors with nuclear weapons will increase until someone makes a mistake. After that, our ability to manage these risks may be destroyed completely.
That said, I also want to share some more insight into the 0.5% figure that I shared at the EA Global meeting. I thought it was important for me to offer the EA community my own assessment about the probability of a catastrophic nuclear event (CNE).
I want to be clear, however, that any probability assessment in this space is, of necessity, a judgment call, no matter who provides it. The nuclear “system” is a massive, complex, system of many sub-systems, and it’s impossible to quantify the risks of use the way one might, for example, conduct a probabilistic risk assessment of the likelihood of failure of a specific engineered technology (where specific data about system performance over time can be employed to calculate probabilities). With the nuclear weapons system, we are talking about thousands of individual weapon and delivery systems, tens of thousands of humans in the decision-making and launch execution loop, hundreds of thousands of technical (and digital) components, and constantly evolving political dynamics and uncertainties which are not quantifiable beyond a qualitative judgment.
Ultimately, I’ve given you my judgment about the probability of nuclear use, on average, in any given year based on:
my experience working within the nuclear system and my understanding of the system design and its vast complexity;
my knowledge of (and in one case direct experience with) failures in the nuclear weapons system over the last 7 decades that led to loss of life or weapons being lost, damaged, or almost used -from numerous operational accidents, incidents, near misses, component failures, faulty warnings, etc (see examples here: www.motherjones.com/politics/2014/11/timeline-nuclear-weapons-accidents-mishaps-near-misses/ and here: www.newyorker.com/news/news-desk/world-war-three-by-mistake);
my understanding and experience of the culture of control of the weapons and its weak points, (for example after the cold war ended, the culture of nuclear control in the US military declined significantly – see a report on this here: https://www.hsdl.org/?abstract&did=486958)
my understanding and experience of global politics and the dynamics of great power conflict.
The key point I wanted to illustrate with my probability assessment is that even if we think the risk of use is small in any given year, the cumulative risk is unacceptably large over time. I did not say, but perhaps should have highlighted, that nuclear risk is not static, and some years my judgment about the risk would be much higher than 0.5%. For example, we are watching Russian forces mass along the border with Ukraine as I write, and tensions are very high right now between the US/Europe and Russia. In this moment of heightened risk, and until we back away from posturing our forces for conflict, I would say the risk of use is much higher, perhaps by as much as a factor of 10. If that persists over time, and between more nuclear adversaries, the probability of use over time becomes much higher. Sadly, I suspect that’s the world we are heading for – an increasing use probability – if we don’t succeed in dramatically reducing risks through a variety of measures that are policy, technical and political in nature.