poly: How to buy shared secrets cheap

Robin Hanson presents an interesting analysis of a game. If one attempts
to apply Robin's insight in the "real world" as Hal Finney suggests then
one should try to take into account other factors such as collusion,
retaliation, and reputation.

>Imagine two people share a secret which would hurt them each $1000 worth
>if it got out. You offer to pay them each $1 to (verifiably) tell you
>their secret.

If the secret holders collude then they may demand $2001 dollars which
they will then split. The slow ramp up of offers will not entice them. If
you only offer a maximum of $2000 they may not reveal the secret at all.
The model Robin presents provides payment to only one secret holder but in
the real world this does not prevent collusion. Also, collusion need not
be explicit. The secret teller may wait for more money with the goal of
trying to placate the non-secret teller with part of the reward or some
other gift of resources if challenged by the non-secret teller at a later
date.

Nick Bostrom mentions the possibility of retaliation. For example, one
secret holder may do another secret holder $5000 dollars worth of damage
if that other person reveals the secret. In this case, revealing the
secret for less than $6000 is not an advantage for the person who would
face retaliation. The secret teller must weigh the cost of retaliation,
the probability of retaliation, the possibility of counter-retaliation.

Another concern is the secret teller's awareness that in the future he or
she faces multiple encounters involving trust. Telling a secret might
damage one's reputation as a secret keeper. This might cause one to lose
future income. The secret teller must try to take into account the
possibility of future lost income.

Robin mentions that:
>The serious problem is that experiments show people deviating from
>standard game theory predictions in similar games.
>The "centipede" game is a famous example...

I wonder if some of the people playing this game are "deviating from
standard game theory predictions" because they are considering factors
such as collusion, retaliation, and reputation. No doubt, the experimental
set-up is intended to provide a simple isolated game but humans typical
engage in repeated encounters in a complex environment. Perhaps the
participants are engaging in "noisy game theory" analysis, or perhaps they
are engaging in an analysis of a different and more global game then the
experimenters intended.

Gregory Sullivan
Received on Mon Mar 2 01:41:34 1998