When should you relocate to mitigate the risk of dying in a nuclear war?

Epistemic Status: “Thinking out loud” is probably a good way to describe it. I’ve been researching and thinking about this stuff on and off for the past few days, and this is the best I’ve got. I haven’t vetted it too closely and probably have made some mistakes.

Some people see me as a very risk-averse person. I wouldn’t say that. It’s more that I’m death-averse. Risks that involve the possibility of death, I’m very averse to, relative to anyone I’ve ever met or can think of off the top of my head. Risks that don’t involve the possibility of death, I think I’m pretty tolerant of. Financially, I start startups and play poker. Not that either of those things are particularly risky. Physically, an example is that I really don’t mind the risks of things like getting mugged or breaking my leg.

I remember the first conversation I had with someone about the Ukraine/​Russia conflict (other than my girlfriend). It was with my mom. I was talking on the phone with her and I think I said something like “You see what’s going on in the Ukraine?”. Her response was something like “Ugh don’t remind me. Gas prices are going to skyrocket.”

That’s not where my mind went. My mind went to death. Nuclear bombs are a thing. Escalation of conflict is a thing. Irrationality is a thing. This very well might end up in a nuclear war that ends up killing me. How high is that risk, and is it worth me doing anything to mitigate it?

My initial instinct was a begrudging yes. I place an extremely high value on life. The risk is actually real. It probably is high enough to justify moving. Which really sucks. I just moved to Portland about a month ago. We’re just setting in here. It’s been amazing. It’s the first time in my life I’ve been able to live in an urban area, which is awesome because I don’t drive. And it’s the first time in my life I’ve gotten to choose my location based on where I want to be, not where I have to be, eg. because of a job.

Next, I ran some quick numbers to get a ballpark of where we’re at and whether I need to act fast. Suppose that we value life at $10B (1-3 orders of magnitude higher than you probably do). Suppose that there is a 110 chance of me dying if the US is attacked and a 0% chance if I move. If there is a 11,000 chance of us getting attacked, staying here would cost $1M. 110,000 it costs $100k. 1100,000 it costs $10k. With that initial analysis, it’s not immediately clear to me what I should do. Which is a success, I think. I wanted to see if an initial analysis yielded an obvious action. It didn’t. Now it’s time for a closer look.

The next thing I did is, after spending a little bit of time googling around for stuff and talking it through a bit with a friend, I wrote RFC WWIII. Writing really helps me think, and I thought it’d be good for the community as well. In that post I dove a little bit deeper into things, but didn’t really make it much further than my initial analysis. Talking some things through in the comments was pretty helpful though.

Then I came across 80,000 Hours’ problem profile on nuclear security. That seems to contain a ton of useful resources and information. I listened to the thing they push the most there, a podcast episode with Daniel Ellsberg.

That significantly changed my view. From what Ellsberg said, it seems quite likely that if there is a nuclear exchange, it will be large enough to trigger a nuclear winter, and if there is a nuclear winter, we’ll all end up dead within a year, save for a few people surviving off of fish and seaweed (and wood?!) in New Zealand, living a Hatchet-style life. If those things are true, moving doesn’t really do much for you. Plus, the reason I value life so highly is because of the possiblity of life extension and an awesome post-singularity future. But in the scenario where I’m living off of fish and seaweed in New Zealand, the probability of life extension is near zero. Plus the quality of that life is quite low.

But then, to continue this roller coaster, I came across a tweet by Rob Wiblin saying that he is leaving London due to the threat of a nuclear exchange.

After listing all the imaginable nuclear escalation scenarios I’ve decided to leave London for now.

The lost life expectancy from remaining narrowly outweighs the inconvenience of leaving for me (which isn’t so big).

To help calibrate, a 1 in 1000 chance = ~2 weeks life lost.

Rob is a guy who’s opinion I have to take seriously. He is the director of research at 80,000 Hours. I only have a limited impression of him, but from what I can tell he seems to have solid epistemics. And he is one of the more informed people on the topic of nuclear security. He did three 3+ hour long interviews of various experts in the field, and seemingly did a bunch of research to prepare for those interviews. And there’s probably more that I’m not aware of. Also, major props to Rob for taking ideas seriously, sticking his neck out, and having the courage and altruism to share his position with the world.

Let’s try to look at this with a focus on where the cruxes are. One crux is how likely it is that an attack leads to a nuclear winter. What is nuclear winter? I’ll let Daniel Ellsberg explain.

Well, that our policy has actually been the threat of an insane action, an action that essentially we now know for the last 35 years has involved killing nearly everyone on earth by the smoke from the burning cities that are planned to be hit in our war plan. And that smoke, we now know on the nuclear winter calculations, would be lofted into the stratosphere, would spread around the world globally. I’m talking now about a war between the U.S. and Russia, where thousands of weapons would be involved. And a few hundred of those weapons on cities which are targeted would be enough to cause smoke that would reduce the sunlight reaching the earth’s surface by about 70%, killing all the harvests worldwide and for a period as a long as a decade.

But that wouldn’t be necessary, killing all the harvests for about a year or even less would exhaust our food supplies, which globally are about 60 days, and nearly everyone would starve to death except for a small fraction, perhaps 1% a little more or less, of humans would survive, in Australia or New Zealand, southern hemisphere is somewhat less affected, eating fish and mollusks. And that could be a sizable number of people. One percent is 70 million people, but 99% gone and virtually all the larger animals other than humans. They’re not as adaptable as we are, and they can’t move thousands of miles and wear clothes, light fires, have houses. They would go extinct altogether, as they did when an asteroid hit the earth 67 or 65 million years ago and created a very similar effect, blotting out the sunlight by the dust that was sent up.

If an attack leads to such a scenario, it looks like I’ll be dead within a year. In which case dying immediately from the attack in Portland vs escaping to Greenland but dying a year later as the crops all fail, it’s not like escaping to Greenland gets me a much better outcome.

Actually, it’s probably a worse outcome. As much as I don’t want to die, I am a believer that there are outcomes worse than death, and a post apocolyptic life in Greenland dreading my inevitable doom as the crops slowly fail, I shiver just thinking about it. It’s probably worse than being dead. So that option isn’t looking very good.

Similar story with moving to New Zealand in hopes of being one of the survivors in this nuclear winter scenario. I feel like there’s a good chance I wouldn’t survive anyway even if I did make it there. In a different 80,000 Hours podcast episode with David Denkenberger, they talked about how New Zealand is probably overrated. There’s a risk that people fight over a limited food supply and things basically get too crazy and don’t work out.

But even if I did survive, that Hatchet-style life doesn’t sound very good. It probably isn’t worse than dying, but it’d be pretty bad. Let’s just say 20% as good as normal life. Suppose we use the typical $10M to value normal life (life extension won’t happen in this scenario). That means this New Zealand life would be worth $2M. And let’s say, generously, there’s only a 50% chance I survive if I make it there. So call it $1M instead of $2M. By moving to New Zealand right now, the logic would be that there is a small chance of losing out on this life valued at $1M. If there is a 11,000 chance of being killed in a nuclear attack here in Portland, then staying here is like taking a 11,000 chance at losing out on $1M. So an EV of -$1,000. Not worth the hassel of moving. Not even close. Even if it was a 1100 chance of an attack, the $10k wouldn’t be worth it to me either. We’d have to start getting into the 110 territory, and I really don’t think we’re there right now, so it’s looking like this is an option that I can shelve for now, and perhaps revisit if things get really crazy.

What a terrible thing to seriously think about though. I really hope I don’t have to.

I did say this was a crux though. So far I’ve been assuming that it is true. But is it? Will a nuclear winter actually happen? Maybe it is only a smaller exchange that doesn’t actually lead to a nuclear winter. Ellsberg doesn’t think so.

So the war plans of both U.S. and Russia have contemplated as sending not just hundreds but thousands of warheads at each other and hitting hundreds of cities. And something between 100 and 200 cities hit that way, by thermonuclear weapons, would cause this nuclear winter. The likelihood of a limited nuclear war between the U.S. and Russia is not quite zero, but it’s very small. Any armed conflict between U.S. and Russia, which has never occurred yet, would bear a high likelihood or a real risk of erupting and escalating into use or nuclear weapons by one or the other. Once that happened, the change of keeping it limited is very low. Each would worry that the other was about to escalate. And another major point in the book is that our planning on both sides has been aimed, delusionally for this entire period, at limiting damage to one’s own side by counterforce, by hitting the forces of the other side in addition to its cities and its urban industrial centers. In fact, most of the targets on both sides are of military targets, many of them near cities or in the cities actually.

He calls it “not quite zero, but it’s very small”. That’s too bad. He is just one guy though. My experience has slowly been teaching me that seemingly smart people can be wrong, and that I trust them too much. I’d be much more comfortable getting more data points on this. If people in the comments can help, that’d be appreciated.

Something that pushes me closer to the possibility of a smaller nuclear exchange that didn’t trigger a nuclear winter is that, it just wouldn’t make sense strategically to have a large exchange. Ellsberg actually talks a lot about this throughout the podcast. You can nuke the crap out of all their land based nuclear launch places and stuff, and even cripple their air force so that they can’t launch nukes by air, but there are still nuclear submarines. It’d be impossible to disarm all of them. So the enemy would still be able to hit you back with their submarines. Hard. So then, if it doesn’t make strategic sense to nuke the crap out of them and start a large exchange, why does Ellsberg think that it is basically inevitable? It’s not quite clear to me, actually.

Yeah, I guess I don’t understand this very well. In which case, I suppose I should assign something like a 20% probablity of a limited exchange.

In this scenario, I personally would be safe in Portland. I think. From the googling I’ve done, it looks like it’d be low enough on the list of targets where it wouldn’t get hit. The targets are places like military bases, critial infrastructure, government buildings like the Pentagon, and major population centers like New York.

Check this out:

It shows the cities, towns and military sites which will be pulverised in the case of a 500-warhead or 2,000-warhead nuclear attack.

The larger attack, which would target 2,000 locations across the USA, would most likely be an unprovoked attack by an enemy.

In this case the enemy would have the element of surprise and would attempt to hit as many big cities and important military sites as possible.

If America attacks first and is hit as response, the enemy would most likely know a victory would be impossible and would simply attempt to kill as many people as possible.

This map, put together by the Federal Emergency Management Agency (FEMA) and the National Resources Defense Council, shows the areas most likely to be attacked in these scenarios.

The purple triangles indicate big cities, while the black circles are smaller cities and towns, as well as military sites and missile launch sites.

Remember, Ellsberg said it would only take a few hundred nukes to trigger a nuclear winter. And Russia would have to attack a bunch of other NATO countries as well, and the various military bases NATO countries have scattered around the globe. So looking at those purple triangles and thinking about all the other places Russia would have to hit, it doesn’t look like Portland would make the cut.

I’m not 100% confident in this though. Maybe I’m… 90% sure? Let’s run some numbers on this. Suppose there’s a 10% chance that in this scenario, Portland gets hit. In this scenario, it’s not guaranteed that I die. I remember reading on some survival blog that something like 50% of people surived in Nagasaki. Looking at these maps of air flow for nuclear fallout, Oregon is one of the best places to be. And looking at this diagram, it looks like if you can manage to get a few miles away from the impact site and stay covered up for 48 hours or so, you’re actually in decent shape. So looking at that 50% number for Nagasaki and factoring in that I’m analytical and paranoid enough to perhaps, whether by car, bike or foot, run away from the city center before an attack, maybe the chance I die is something like 20%?

So, we have a 20% chance of a smaller scale attack rather than a larger scale one, 10% chance that Portland gets hit in this scenario, and a 20% chance I die if Portland gets hit. Suppose that in this smaller scale scenario we value life at $2B instead of that $10B I default to, because life extension is less likely. If there is a 11,000 chance of an attack in general, then we have an EV of 0.1% * 20% * 10% * 20% * $2B = -$8k.

So, would I live in, say, rural Oregon for a year in exchange for $8k? No, I wouldn’t. I like where I’m at. What about for $80k, ie if there was a 1100 risk? Yeah, I would. So 1100 seems like the right order of magnitude to target for when I should consider a move. There’s also the idea of getting an Airbnb somewhere for a week or so if tensions are particularly high.

What about for people in places like New York? Suppose New York has something like a 80% chance of getting hit instead of Portland’s 10% chance. And instead of a 20% chance of dying, it’s 40%, because of air flow and nuclear fallout. That’s a 8 * 2 = 16 times bigger risk. So I suppose 1/​1000 chance of an attack is the order of magnitude I’d target if I lived there.

Let’s return to that tweet by Rob Wiblin. He did mention that he doesn’t buy the idea of there being no chance of surviving long term, and linked to two podcast episodes he did: one with David Denkenberger and the other with Luisa Rodriguez. Let’s explore those and see what we find.

The title of Denkenberger’s podcast is “We could feed all 8 billion people through a nuclear winter. Dr David Denkenberger is working to make it practical.” The opening quote is:

I’m very concerned that if people don’t know about resilient foods then they could conclude that most people are going to die.

It could be an incentive for countries to do very bad things, like steal food from your neighboring countries.

That’s why I want to get the message out that we could actually feed everyone if we cooperate.

The opening paragraph is:

If there’s a nuclear war followed by nuclear winter, and the sun is blocked out for years, most of us are going to starve, right? Well, currently, probably we would, because humanity hasn’t done much to prevent it. But it turns out that an ounce of forethought might be enough for most people to get the calories they need to survive, even in a future as grim as that one.

The podcast is three hours long and I haven’t listened to the whole thing (or read the whole transcript), but it sounds like he’s saying that we could do things to prevent starvation in the event of a nuclear winter. I’m not seeing anything about, as things are today, it being likely that we would avoid starvation. In fact, in that opening paragraph, it says that most of us would starve. And looking through the rest of the intro, it seems that if we were to survive, it would mean getting creative with things like mushrooms. That sounds like a pretty bad life to me.

So this isn’t helping my estimate of survival in a nuclear winter. In fact, it’s hurting it. If nuclear winter was more survivable, I would not expect to see this guy dedicate his career to this and make all of these claims about creative ways to farm mushrooms and the need to extract sugar from wood. I take this as pretty strong evidence that a nuclear winter would be quite hard to survive (which I mostly factored in to my probability estimates above actually, I’m just explaining it in the post here).

There was a section in the podcast that the transcript titled “Should listeners be doing anything to prepare for possible disasters?”. I was confused by it. Denkenberger brings up the idea of avoiding popular cities. But he isn’t very enthusiastic about it. He does say:

It really depends on how many nuclear weapons are used and whether it’s just the US as the target or all of NATO. But I think that just living on the outskirts of a city is quite a bit lower risk.

But what about nuclear winter? Isn’t his whole thing that, as things currently stand, there wouldn’t be enough food for people in the event of a nuclear winter, in which case it doesn’t matter where you’re located? So living on the outskirts would only help in the case of a smaller scale attack. To be charitable, I’ll assume that’s what he had in mind. I wish he would have mentioned and elaborated on that though. It’s a very important question how plausible this smaller scale attack is. Ellsberg called it a “not quite zero” chance. If this is what Denkenberger dedicated his career to, I’d think he’d have spent quite a bit of time thinking about this question. So I’m disappointed to not see more discussion of it.

Let’s move to the episode with Luisa Rodriguez. The title of that one is “Luisa Rodriguez on why global catastrophes seem unlikely to kill us all”. And yeah, that’s what the podcast is about. The idea that not every human would die, and that humanity would slowly build itself back up. Ie. we’d be back to where we are now within 1,000 years.

I don’t see how that is relevant though. Or, maybe I should say helpful. I don’t see how it is helpful. It sounds like more evidence that we would in fact have almost all of humanity die in a nuclear winter, and the ones who surive would be grasping at straws.

But Luisa has a series of articles on the Effective Altruism forum. Cool! Let’s see if there’s anything good there.

Opens articles. Skims through them. Wow! The roller coaster continues! Another significant belief update.

Aparently a lot of EAs and people in general just assume that a US-Russia nuclear exchange would result in this nuclear winter we’ve been talking about that would kill almost everyone. But Luisa doesn’t believe that, and her reasoning seems sound.

Let’s see what her estimate of an only 11% chance of a nuclear winter would mean. Actually, let’s kinda split the difference and assume a 20% chance. She seems more optimistic than other people, and a higher number helps account for the fact that living in a world where a nuclear attack happened but it didn’t trigger a nuclear winter wouldn’t be as good a world to be a part of. Plus, it’s seeming that recent events are evidence against Putin being rational, which I think means a larger exchange is more likely.

Previously I said the following:

So, we have a 20% chance of a smaller scale attack rather than a larger scale one, 10% chance that Portland gets hit in this scenario, and a 20% chance I die if Portland gets hit. Suppose that in this smaller scale scenario we value life at $2B instead of that $10B I default to, because life extension is less likely. If there is a 11,000 chance of an attack in general, then we have an EV of 0.1% * 20% * 10% * 20% * $2B = -$8k.

Now we’re saying an 80% chance it’s a smaller scale attack. So it’s 0.1% * 80% * 10% * 20% * $2B = -$32k. Huh. I guess that doesn’t change things too much. I previously was giving a 20% chance of a smaller scale attack even though Ellsberg said the chance was “not quite zero” because I’ve learned not to trust people. Pats self on the back. Thanks Professor Quirrell.

By the way, I’m feeling much better about these probabilities now. Getting another data point is great. Various cool people in the EA community reviewed Rodriguez’s work, so it’s implicitly even more data points. I like that I’ve seen, presumably, people on both far ends of the spectrum. Ie. Ellsberg seems close to the end of “it’s gonna be a nuclear winter and everyone is gonna die” whereas Rodriguez seems close to the opposite end.

Let’s try another adjustment. Rodriguez leans against thinking that an exchange would target cities and instead it’d be military bases. Previously I was assuming a 10% chance that Portland would be hit in this small scale exchange scenario. If feel like I can update away from that. Maybe to 5% instead? Sure. That gives us -$16k. Still in the same vicinity.

I could continue diving into this, but I think this is a good place to stop. Orders of magnitude are what matter. These smaller changes don’t matter too much. It seems that a roughly 1100 chance of an attack is where it’d make sense to relocate for someone like me in a city like Portland. For someone in a city like New York, 11,000 seems more appropriate.

Edit: That assumes the super high $2B value on life. Other people probably value it 1-3 orders of magnitude less. I probably shouldn’t have used such a high value for myself either. I think I overestimated/​didn’t really think about the chance of an aligned AI happening in this world where a smaller scale nuclear attack happens. In such a world, it’s possible that technological progress is halted, but perhaps more concerningly, I think I’d expect it to be more likely for governments to seek to use AI as a weapon, increasing the chance that it is unaligned. I’ll guess that this adds up to a 10x reduction in the value I place on life, so $200M instead of $2B, in which case it’d be 110 for Portland, but I could easily be way off.