[EMAIL PROTECTED]">More precisely, you cannot fool the system into giving you better results, butIf I understand correctly, this tactic will neither help nor hinder our
friends. It will not hinder our friends because the method is monotonic.
It will not help our friends, because you cannot fool the system by
putting in insincere utilities.
you could always fool it into giving you worse results (e.g., by reversing some
of your preferences). But yes, I think your statement is correct with that
modification.
[EMAIL PROTECTED]">I'm not sure. If E and G are front runners, and you like E better, then probablyAnother thing. It seems to me that if Cranor advised me to approve AB
but the election went to E (skipping over C and D), I would be pretty
upset, especially if it turned out that everyone voted precisely as
advised by Cranor.
Therefore I assume that once Cranor figures out that E is going to win,
she is going to advise me approve down to E.
so. I think the central program needs to be making a probabilisitic rather than
a deterministic prediction. One way it could do this is by only statistically
sampling the inputs to make the next prediction. Then it never "figures out that
E is going to win", or at least if it does it doesn't feed this fact back to the
individual strategizers. All the individual strategizers should see is a matrix
of probabilities calculated by the central algorithm.
[EMAIL PROTECTED]">I agree with that analysis.Whether I am advised to
approve E itself or not would depend on whether my (faction's ?) failure
to support E would result in someone I considered below E winning.
Suppose that E is destined to win (if everyone votes their optimum
strategy) and that everyone who is advised to vote down to and including E
follows that advice, but that the other folks rebel and refuse to vote
down to just above E. Then E would still have to win, because no other
candidate increases in approval.
On the other hand, if some of the "including E" voters foolishly try to fool
the system, it could backfire on them. They will either increase support for
another candidate, or reduce support for E, or have no effect at all. Rarely
would this result in a candidate they like better than E getting elected, unless
the system's original prediction about E was inaccurate in the first place. If
a enough of these voters actually preferred C to E for this strategy to work,
then it seems unlikely the system would have projected E in the first place.
Richard
[EMAIL PROTECTED]">
