I think thereâs a good chance this basic point is right, but Iâm not sure your takeaway from the Samotsvety forecast is correct? I think the 3-100 hours lost in expectation is based on the current information about risk. The Samotsvety forecast is that conditional on a nuclear weapon being used in Ukraine, there is a ~2% chance of London being nuked. I think the mean estimate for expected hours of life loss if one stays in London in that case is ~2000. Thatâs a substantial number of lost hours, and I can see it being rational to get to a safer location if those are the stakes.
Youâre right, my post doesnât make clear enough the difference between current risk and risk conditional on nuclear use in Ukraine.
Trying to figure out expected hours lost in the latter case seems to depend a ton on which of their forecasts you look at. My instinctive reaction was that 2000 is way too high, as theyâre at 16% on Russia using a nuclear weapon in Ukraine so it can only increase risk by a factor of 6 or so if it happens, but they state itâd raise risk by a factor of 10 or so if it happened. Iâm going to use the factor of 6 because I donât understand how they got 10 and it reads like it might just be an order of magnitude estimate.
Using their âforecastersâ aggregateâ, where the mean is 13, the hours lost conditional on use in Ukraine is still less than 100 hours. Using their âfull rangeâ, where the mean is 150, the hours lost conditional on use in Ukraine is 1000. That suggests itâs quite important to figure out which of those aggregating methods make more sense, as I suspect the costs of fleeing are generally higher than 100 but less than 1000 hours. (Though fleeing in the least costly way could reduce the costs of fleeing enough to be less than 100 hours and thus worth it even in the lower case.)
Hmm interesting, I got 2000 by just setting rusiaUsesNuclearWeaponsInUkraine to 1 in the squiggle model. Looking at it further, the mean moves around between runs if I just use 1000 samples. Updating to 1000000, it seems to converge on 1700.
I agree that this is a place where forecast aggregation adds a lot of challenges.
This is also tricky because I donât think it lets you compare to the option Iâd actually advocate for, which is something like âflee at a slightly later pointââthe US has good intel on Russia, and it seems likely that US officials will know if Russia appears to be headed towards nuclear war. If you have to compare âflee the instant a tactical nuke is used in Ukraineâ or âstay no matter whatâ, âstay no matter whatâ doesnât look good, but what you want to compare is âflee the instant a tactical nuke is used in Ukraineâ to âflee at some subsequent sign of dangerââthat is, the real question is how many life-hours you get by fleeing early that you donât get by fleeing late (either because we donât get any warning, or because by then many people are panicking and fleeing).
There are a bunch of preparations the US military would want to take in the face of elevated odds of nuclear war (bombers in the air, ships looking for submarines, changes of force concentration) and I donât believe they will sacrifice making those preparations for crowd management reasons. I agree itâs possible theyâll say something noncommittal or false while visibly changing force deployments to DEFCON 2 or whatever, though this is not what they did during the Cold War and it would be pretty obvious.
My impression is that US intelligence has been very impressive with regard to Russiaâs military plans to date. US officials confidently called the war in Ukraine by December and knew the details of the planned Russian offensive. Theyâre saying now that they think Putin is not imminently planning to use a tactical nuke. If theyâre wrong and Putin uses a tactical nuke next week, thatâd be a big update they also wonât predict further nuclear escalation correctly, but my model is that before the use of a tactical nuke, weâll get US officials saying âweâre worried Russia plans to use a tactical nukeâ. If Iâm right about that, then I further predict theyâll be giving pretty accurate assessments of whether Russia is going to escalate from there.
That suggests a threshold to leave of [ tactical nuke use in Ukraine, if it surprises US officials] or [after tactical nuke use in Ukraine and a warning from US officials that Putin seems inclined to escalate further after tactical nuke use], which would be a 10x or more further update on risk in my view.
Though I should say that I think tac nuke use in Ukraine is also a reasonable trigger to leave, depending on your personal situation, productivity, ease of leaving, where youâre going, etcâI really just want people to be sure they are doing the EV calculations and not treating risk-minimization as the sudden controlling priority.
How far in advance would you expect US officials to warn the public of the possibility of nukes? (i.e. how much time would we have between such a warning and needing to have left already?)
I donât know, but I think likely days not weeks. Tactical nuke use will be a good test ground for thisâdo we get advance warning from US officials about that? How much advance warning?
Your expected life hours lost become Remaining life hours * P(nuke in your location | nuke in Ukraine), if Ukraine is hit and you choose to stay in your location afterwards. While the multiplier does depend on P(nuke in Ukraine), P(nuke in your location | nuke in Ukraine) is still more important since your location is what determines whether it swings you over the decision threshold or not.
In this framework, before the tac nuke use in Ukraine, your expected life hours lost was remaining life hours*P(nuke in your location | nuke in Ukraine) * P (nuke in Ukraine), so your subsequent expected life hours last should change by a factor of 1/âP(nuke in ukraine), or about six.
Though I think straightforwardly applying that framework is wrong, because it assumes that if you donât flee as soon as thereâs nuke use in Ukraine, you donât flee at all even at subsequent stages of escalation; instead, you want P(nuke in your location| nuke in Ukraine and no later signs of danger which prompt you to flee). To figure out your actual expected costs from not fleeing as soon as thereâs tactical nuke use in Ukraine, you need to have an estimate of how likely it is that thereâd be some warning after the tactical nuke use before a nuclear war started.
Yes, it does rely on that simplified assumption. I think Iâm unlikely to get more than 1 additional bit of information via further warnings after a nuke in Ukraine (if that), so staying doesnât seem worth the risk, but if you think you get legible warning signs >84% of the time (or whatever 1 - p(nuke in Ukraine) is) then it seems worth waiting.
ETA: to clarify, my general position is that while Iâm open to the possibility that thereâll be further signals which convey more bits of information about which world youâre in than the initial ânuke in Ukraineâ signal, I expect those extra bits wonât do me much good because in most of those worlds events will move fast enough that I wonât be able to usefully respond. If you have a lot of weight on âescalation, if any, will be slowâ, then your calculation will look different.
I think thereâs a good chance this basic point is right, but Iâm not sure your takeaway from the Samotsvety forecast is correct? I think the 3-100 hours lost in expectation is based on the current information about risk. The Samotsvety forecast is that conditional on a nuclear weapon being used in Ukraine, there is a ~2% chance of London being nuked. I think the mean estimate for expected hours of life loss if one stays in London in that case is ~2000. Thatâs a substantial number of lost hours, and I can see it being rational to get to a safer location if those are the stakes.
Youâre right, my post doesnât make clear enough the difference between current risk and risk conditional on nuclear use in Ukraine.
Trying to figure out expected hours lost in the latter case seems to depend a ton on which of their forecasts you look at. My instinctive reaction was that 2000 is way too high, as theyâre at 16% on Russia using a nuclear weapon in Ukraine so it can only increase risk by a factor of 6 or so if it happens, but they state itâd raise risk by a factor of 10 or so if it happened. Iâm going to use the factor of 6 because I donât understand how they got 10 and it reads like it might just be an order of magnitude estimate.
Using their âforecastersâ aggregateâ, where the mean is 13, the hours lost conditional on use in Ukraine is still less than 100 hours. Using their âfull rangeâ, where the mean is 150, the hours lost conditional on use in Ukraine is 1000. That suggests itâs quite important to figure out which of those aggregating methods make more sense, as I suspect the costs of fleeing are generally higher than 100 but less than 1000 hours. (Though fleeing in the least costly way could reduce the costs of fleeing enough to be less than 100 hours and thus worth it even in the lower case.)
Hmm interesting, I got 2000 by just setting rusiaUsesNuclearWeaponsInUkraine to 1 in the squiggle model. Looking at it further, the mean moves around between runs if I just use 1000 samples. Updating to 1000000, it seems to converge on 1700.
I agree that this is a place where forecast aggregation adds a lot of challenges.
This is also tricky because I donât think it lets you compare to the option Iâd actually advocate for, which is something like âflee at a slightly later pointââthe US has good intel on Russia, and it seems likely that US officials will know if Russia appears to be headed towards nuclear war. If you have to compare âflee the instant a tactical nuke is used in Ukraineâ or âstay no matter whatâ, âstay no matter whatâ doesnât look good, but what you want to compare is âflee the instant a tactical nuke is used in Ukraineâ to âflee at some subsequent sign of dangerââthat is, the real question is how many life-hours you get by fleeing early that you donât get by fleeing late (either because we donât get any warning, or because by then many people are panicking and fleeing).
Why would you think that they would transmit this information honestly, rather than managing the crowd and try to have people not panic?
There are a bunch of preparations the US military would want to take in the face of elevated odds of nuclear war (bombers in the air, ships looking for submarines, changes of force concentration) and I donât believe they will sacrifice making those preparations for crowd management reasons. I agree itâs possible theyâll say something noncommittal or false while visibly changing force deployments to DEFCON 2 or whatever, though this is not what they did during the Cold War and it would be pretty obvious.
Cheers.
(not sure how to interpret this)
Iâd expect it to be harder to tell that Russia is heading towards nuclear war than that they are planning an invasion.
That makes sense to me, I agree thatâs a good relevant comparison.
Thanks for writing this! Do you have a particular sign of danger in mind? I donât feel that I would know what else to look for as a leave trigger.
My impression is that US intelligence has been very impressive with regard to Russiaâs military plans to date. US officials confidently called the war in Ukraine by December and knew the details of the planned Russian offensive. Theyâre saying now that they think Putin is not imminently planning to use a tactical nuke. If theyâre wrong and Putin uses a tactical nuke next week, thatâd be a big update they also wonât predict further nuclear escalation correctly, but my model is that before the use of a tactical nuke, weâll get US officials saying âweâre worried Russia plans to use a tactical nukeâ. If Iâm right about that, then I further predict theyâll be giving pretty accurate assessments of whether Russia is going to escalate from there.
That suggests a threshold to leave of [ tactical nuke use in Ukraine, if it surprises US officials] or [after tactical nuke use in Ukraine and a warning from US officials that Putin seems inclined to escalate further after tactical nuke use], which would be a 10x or more further update on risk in my view.
Though I should say that I think tac nuke use in Ukraine is also a reasonable trigger to leave, depending on your personal situation, productivity, ease of leaving, where youâre going, etcâI really just want people to be sure they are doing the EV calculations and not treating risk-minimization as the sudden controlling priority.
How far in advance would you expect US officials to warn the public of the possibility of nukes? (i.e. how much time would we have between such a warning and needing to have left already?)
I donât know, but I think likely days not weeks. Tactical nuke use will be a good test ground for thisâdo we get advance warning from US officials about that? How much advance warning?
Your expected life hours lost become
Remaining life hours * P(nuke in your location | nuke in Ukraine), if Ukraine is hit and you choose to stay in your location afterwards. While the multiplier does depend onP(nuke in Ukraine),P(nuke in your location | nuke in Ukraine)is still more important since your location is what determines whether it swings you over the decision threshold or not.In this framework, before the tac nuke use in Ukraine, your expected life hours lost was remaining life hours*P(nuke in your location | nuke in Ukraine) * P (nuke in Ukraine), so your subsequent expected life hours last should change by a factor of 1/âP(nuke in ukraine), or about six.
Though I think straightforwardly applying that framework is wrong, because it assumes that if you donât flee as soon as thereâs nuke use in Ukraine, you donât flee at all even at subsequent stages of escalation; instead, you want P(nuke in your location| nuke in Ukraine and no later signs of danger which prompt you to flee). To figure out your actual expected costs from not fleeing as soon as thereâs tactical nuke use in Ukraine, you need to have an estimate of how likely it is that thereâd be some warning after the tactical nuke use before a nuclear war started.
Yes, it does rely on that simplified assumption. I think Iâm unlikely to get more than 1 additional bit of information via further warnings after a nuke in Ukraine (if that), so staying doesnât seem worth the risk, but if you think you get legible warning signs >84% of the time (or whatever
1 - p(nuke in Ukraine)is) then it seems worth waiting.ETA: to clarify, my general position is that while Iâm open to the possibility that thereâll be further signals which convey more bits of information about which world youâre in than the initial ânuke in Ukraineâ signal, I expect those extra bits wonât do me much good because in most of those worlds events will move fast enough that I wonât be able to usefully respond. If you have a lot of weight on âescalation, if any, will be slowâ, then your calculation will look different.