From: Wei Dai (weidai@weidai.com)
Date: Thu May 08 2003 - 18:07:55 MDT
Suppose you are a benevolent dictator and you just obtained intelligence
that suggest a particular person may be planning a suicide attack. You
believe that the intelligence is correct with probability p, and if so
there is a probability q that the attack will succeed and kill an expected
N people. So the expected number of dead people if you do nothing is
pqN.
How high do p, q, and N have to be before you decide to arrest the
person? Assuming you are a utilitarian, the answer is probably something
like "if pqN > C", where C is the cost of arresting an innocent person
divided by the cost of a death.
Unfortunately, most governments are not benevolent dictatorships, so this
answer doesn't work for them. The main problem is that if N is very large,
the rule says to arrest the suspected terrorist even if p is small, which
makes the system easy to abuse. Those in power can falsely accuse their
enemies of terrorism knowing that it's virtually impossible for a
defendant to establish with certainty that he is innocent.
The rule that ends up being applied in practice is "if p > r" where r is
some constant on the order of 1, which is obviously highly suboptimal,
especially in cases where N is large.
Here's where Robin Hanson's futarchy idea (see his post earlier today)
comes in. Establish betting markets, three for each person, on whether
that person will attempt an attack, whether he will succeed, and how many
casualties the attack will cause. Then apply a rule based on the asset
prices in these markets. Assuming that these markets cannot be easily
manipulated, this should be a much more efficient way to deal with
terrorism.
This archive was generated by hypermail 2.1.5 : Thu May 08 2003 - 18:19:01 MDT