This site simulates the dynamics of rational deliberation as
described by Brian Skyrms (for
instance here). Given
a range of options and an initial probability distribution over these
options (a state of indecision), each deliberation step computes a new
probability distribution by privileging options with greater
expected utility.
More precisely, the update proceeds by Nash Dynamics: let
the virtual utility VU of a state of indecision P be the
expectation VU(P) = sum_i U(A_i)*P(A_i). If the expected utility of an
action A in a state of indecision P is greater than the virtual
utility of P, the difference is called the covetability of A:
Cov(A) = max(U(A)-VU(P), 0). Nash Dynamics updates the probabilities P
to P'(A) = (k*P(A) + Cov(A)) / (k + sum_i Cov(A_i)), where k is a
constant determining how slow the decision maker moves in the
direction of actions that seem better (the higher, the slower). k is
currently set to . Try
increasing it if you don't reach an equilibrium.
Examples:
Death in Damascus (from Gibbard & Harper 1978)
Newcomb's Centipede (from Brian Weatherson 2008)
-- wo@lalalaumsu.de 2008.