[ExI] Conspiracy epistemology (Was: cut off worried)
Anders Sandberg
anders at aleph.se
Fri Feb 19 01:21:53 UTC 2016
On 2016-02-19 00:38, Rafal Smigrodzki wrote:
> On the other hand, what if the conspiracy *wanted* to make it look
> like an assassination? Or maybe the opposite - What if a disgruntled
> hotel maid wanted to take the political future of the country in her
> own hands but looked into the dead eyes of the victim and just had to
> cover them up?
>
> Mind boggles. Shadows behind every pillow.
If you have the brainpower you can of course run Bayes for each
hypothesis: P(assassination|evidence), P(fake assassination|evidence),
P(murder looking like assassination|evidence)... All have fairly low
priors.
As the hypotheses proliferate, the denominator gets lots of terms of the
form P(E|Hi)P(Hi). It seems a pretty safe bet to argue that the
hypothesis no foul play has a dominant prior, so again the factor will
look like P(evidence|hypothesis)/P(evidence|no foul play)P(no foul
play). Just because we can make the list of hypotheses arbitrarily long
doesn't mean it steals enough probability from the no foul play prior
(remember, that one is about 0.99996, while most complex theories are
the product of 4e-5 with something smaller) to make the Bayes factor large.
Conspiracy theory seems to be based on finding a hypothesis where
P(evidence|hypothesis) is really high, and then disregard the low prior
for the hypothesis. Often Bayes factors are then mis-evaluated by noting
further evidence that fits (the numerator) but ignoring base rates and
how it fits to no foul play (denominators). When people point out the
denominators, one invokes a few extra possibilities to make it seem that
they are smaller.
Meanwhile the muppet ninja sent by Ernie and Bert to help Sotomayor (she
befriended them when she was on Seasame Street) is quietly acting like a
normal pillow. P(pillow|ninja)/P(pillow|no foul play)P(no foul play) =
1. But we know better, since the rubber ducky told us.
--
Anders Sandberg
Future of Humanity Institute
Oxford Martin School
Oxford University
More information about the extropy-chat
mailing list