[ExI] Conspiracy epistemology (Was: cut off worried)

Anders Sandberg anders at aleph.se
Thu Feb 18 23:28:42 UTC 2016

On 2016-02-18 15:35, spike wrote:
> ... By the way, does anybody have a good long-term database of 
> historical assassinations?
> -- 
> Anders
> We should have a way of dealing with those cases where foul play is un likely but possible.  We might estimate perhaps 10 percent chance Scalia was slain, ja?

Actually, way lower. Think of it like this: the murder rate of the US is 
about 4 per 100,000, or a probability of 0.00004.

There are no doubt some murders that are successfully hidden as natural; 
if we assume for every discovered murder there is a hidden one we are 
likely extremely paranoid, but that brings up the probability to 0.00008.

Now, assassinations are actually fairly rare: most murder is of 
non-prominent persons. Even there, judges are uncommon (in the data in 
Snitch, T. H. (1982). Terrorism and political assassinations: A 
transnational assessment, 1968-80. /The Annals of the American Academy 
of Political and Social Science/, 463(1), 54-68. only 3% of successful 
assassinations involved judges). So thinking a particular judge's death 
is due to foul play without any supporting evidence should lower the 
probability of the statement being true by quite a lot - a very rough 
assassination rate (assassinations/deaths) I estimated using Wikipedia 
statistics is 7.8e-4. A lot of those deaths were murders, so let's be 
charitable and say a factor of 1e-3. That makes a probability of 8e-8.

So, the ultra-paranoid prior for Scalia being assassinated is a 
probability of 8e-8.

Now, if you started to get positive evidence that prior would be 
multiplied by Bayes factors of the form 
P(evidence|assassination)/(P(evidence|assassination)P(assassination) + 
P(evidence|no assassination)P(no assassination), which is roughly, since 
assassinations are rare, P(evidence|assassination)/P(evidence|no 
assassination). For example, lack of autopsy is pretty likely in either 
case, so the Bayes factor is only slightly larger than 1. Seeing ninjas 
disappear from the scene produces a big factor, since they rarely show 
up when people die for non-assassination reasons. Not seeing ninjas 
(because maybe they are invisible like all really good ninjas!) produces 
a Bayes factor of 1: you would expect that in either case.

So, can anybody scrounge up enough Bayes factors to get the prior up to 0.1?

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/20160218/f93e20be/attachment.html>

More information about the extropy-chat mailing list