This paper finds support for the evolution of interdependent preferences un
der natural selection from a (perhaps) surprising source: decision making u
nder uncertainty. Individuals choose from sets of risky alternatives. The l
otteries may involve either idiosyncratic risk or aggregate uncertainty or
both. Robson (J. Econ. Theory 68 (1996), 397-324) gives the evaluation crit
erion for lotteries that maximizes reproductive value and shows that it doe
s not satisfy the expected utility theorem. Cooper and Kaplan (J. Theoret.
Biol. 94 (1982). 135-151) have demonstrated that when lotteries are aggrega
te. the optimal decision rule involves randomization. This paper reexamines
Robson's evaluation criterion. using it to solve fur the optimal amount of
randomization. The solution is characterized and an interpretation is offe
red that links maximizing reproductive value to maximizing expected relativ
e offspring. It is shown that when agents preferences arise from their own
offspring relative to the average within the population, a game is construc
ted out of the choice of lotteries and that the unique Nash equilibrium dis
tribution of actions maximizes reproductive value. (C) 2001 Academic Press.