[ExI] Nuclear probabilities (Was:] Repudiating the national debt)

Anders Sandberg anders at aleph.se
Mon May 16 10:10:20 UTC 2016

On 2016-05-16 07:03, Dan TheBookMan wrote:
> On Fri, May 13, 2016 at 12:37 AM, Anders Sandberg <anders at aleph.se 
> <mailto:anders at aleph.se>> wrote:
> > On 2016-05-12 21:41, Dan TheBookMan wrote:
> > "While I wouldn't want to risk it, my guess is other folks in the 
> chain of command
> > would likely not follow orders. However, let's set that aside. Let's 
> say you're right:
> > Trump in office would increase the odds of a nuclear war. By how 
> much? Maybe
> > Caplan is right about the overall 2.5 times risk. Let's say 2.5 
> times whatever the
> > base rate would be or, better, than Clinton or Sanders. (My guess is 
> Sanders
> > would be less bellicose than either Trump or Clinton.)
> >
> > You can estimate the base rate by doing a Bayesian update on a 
> uniform prior
> > [0,1] of nuclear war probability per year, given 70 years of no war. 
> That gives
> > you an expected 1.4% risk per year.
> >
> > If we accept the 2.5 increase, that means 3.5% risk per year. Over 4 
> years that
> > is 13% risk of a nuclear war (compared to 5.4% for normal presidents).
> I think that estimate is too high. Caplan's view is for a major war -- 
> not necessarily a nuclear war. I think we'd have to include another 
> term to determine what the rate of any major war turning into a 
> nuclear one. But let's run with it.

One can use Martin Hellman's Markov chain model: you have states of 
[peace], [international crisis], [major conflict] and [nuclear war], and 
transition probabilities between them.

How many years of international conflict have we had? If we use the 20th 
century and make it all UScentric we had at least two conflicts (WW I 
and WW II), lasting in total 9 years and with a transition risk to peace 
of about 0.2 per year. One might add the Korean War, where nuclear 
attacks were considered. I think assuming a 1/4 risk of nuclear attack 
in a direct power conflict is not crazy given WW II, but lets say 1/5.

Not certain how many years should be regarded as international crisis 
years. Looking at
suggests that the US was involved in the Berlin Blockade, the Cuban 
Missile Crisis, the Pueblo Incident, the 1973 Chilean coup, Ace murder 
incident, the Iran hostage crisis, and Able Archer. These, plus the two 
major wars, suggests a probability of about 0.09 per year of a crisis 
happening, and then about 0.2 risk of war. We had at least 2-3 crisis 
situations that nearly caused nuclear war, so that transition arrow from 
crisis to war is about 0.02 or 0.03. We also had two cases of internaiton

peace -> crisis 0.09
peace -> peace 0.91
crisis -> conflict 0.2
crisis -> nuclear 0.02
crisis -> peace 0.78
conflict -> nuclear 0.2
conflict -> peace 0.2
conflict -> conflict 0.6

The transition matrix becomes:

[0.91   0.09   0      0

  0.78   0      0.2    0.02

  0.2    0      0.6    0.2

  1      0      0      0]

Note the debatable model that peace occurs after nuclear wars (the last 
line is [1 0 0 0]). In any case, the model predicts the following 
distribution: peace 87% of the years, crisis 7.8%, war 3.9% and nuclear 
war 0.93%.

One can also make nuclear wars a final state, [0 0 0 1]. I have a paper 
(in writing limbo, unfortunately) on the correction factors due to 
anthropic selection - if there are no survivors, then you cannot ever 
observe a nuclear war in your past, so the transition probabilities 
would be biased in a nontrivial way. It can be corrected, and then 
near-misses like the Petrov incident give some information. Right now 
the conclusions look rather reassuring, but that might simply be because 
nuclear disasters are somewhat survivable (if they just kill half of the 
observers you get a 2/3 bias).

If one buys the xrisk model instead, the time spent in peace in this 
post's model is 88%, crisis 7.9%, and war 3.9% - in this particular case 
there is actually very little change.

This model also supports the conclusion that the nuclear war risk per 
year is on the order of a percent.

Clearly real-world data gives us reasons to update the current level of 
concern in different ways, but this represents a large scale "outside 
view" of a century. Obviously one can tinker with the model and 
especially the transition probabilities endlessly. But I think the 
results will not be astronomically different.

> > Now, what can you do about this? Panic? Build a bomb shelter? My 
> guess is
> > very little aside from get worked up."
> >
> > Move to Tasmania?
> Bouvet might be better. :/

4 days of sunshine per year? I will take Tasmania any time. Even Kerguelen.

(My dad had a minor obsession with the island, and there is actually a 
map of it hanging in my room. Mostly ice, with the volcanic beaches 
inhabited by penguins and seals. Not the most hospitable place. )

Anders Sandberg
Future of Humanity Institute
Oxford Martin School
Oxford University

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20160516/44848f97/attachment.html>

More information about the extropy-chat mailing list