Learning to forage in a variable environment

Abstract

In a stochastic environment animals should and really do make their foraging decisions not only according to the expected net benefits but also according to the associated variances. In this study I show how the RPS (relative payoff sum)-learning rule ( Harley, 1981 ) deals with the problem of choosing between two food patches where the expected net benefits are equal but the associated variances differ. The RPS-learning rule is a stochastic rule in which the probability of choosing a patch is proportionate to the relative payoff that the respective patch has yielded so far. To account for a decay of memory in time the RPS-rule gives more weight to recent payoffs. By using two different simulation procedures I show that the RPS-learning rule is either indifferent between constant and variable rewards or develops a preference for constant over variable rewards depending on the foraging regime simulated. The RPS-learning rule therefore represents a mechanism to produce risk-neutral or risk-averse foraging preferences. It is discussed how this result could come about and what it means in relation to empirical findings in a number of animal species.