EPICURE: An Agent-Based Foraging Model Michael E. Roberts, Robert L. Goldstone Department of Psychological and Brain Sciences Indiana University robertsm@indiana.edu Abstract We present an agent-based foraging model, EPICURE, which captures the results from recent human group foraging experiments (Goldstone and Ashpole, 2004; Goldstone et al., 2005), provides a novel explanation for those results and previous animal foraging results, and makes predictions for future foraging experiments. We describe a series of simulations that test the sources of resource undermatching often found in group foraging experiments. We conclude that foraging group size, food rate, and spatial distribution of food interact to produce undermatching, and occasionally, overmatching, to resources. Furthermore, we present wealth distribution results from the aforementioned empirical studies and EPICURE simulations. Introduction Animals often forage for resources and even mates in groups. By congregating with others, individuals can acquire social sampling information and learn new strategies in order to improve food intake and mate selection rates, but these advantages can be compromised by member competition and density-dependent interference and perceptual limitations. In this paper, we describe an agent-based model that captures the interplay between individuals’ foraging strategies and the emergent group foraging behavior. Group foraging distributions are often compared to the ideal free distribution (IFD) model (Fretwell & Lucas, 1970), which predicts that a group of foragers will distribute themselves to resource patches in proportion to the relative resources available at each patch. In an environment where one resource pool holds 80% of the resources while a second pool holds the remaining 20% of resources, the IFD predicts that a group of foragers will optimally distribute themselves to the resource pools, with 80% of the foragers in the first pool and 20% in the second pool. However, many experiments report systematic undermatching in which fewer than expected foragers attend the more profitable patch while more than expected foragers attend the less profitable patch. Undermatching has been found in cichlid fish (Godin and Keenleyside, 1984), zebrafish (Gillis and Kramer, 1984), pigeons (Baum and Kraft, 1998), and humans (Goldstone et al., 2005). In a meta-analysis of undermatching in animal foraging studies, Kennedy and Gray (1993) conclude that information regarding the “relative and absolute resource availability, number of animals, perceptual abilities of animals, competitive interactions, competitive abilities of animals, and the effects of travel between sites” (p. 165) may all lead to undermatching and violate the IFD. Goldstone and Ashpole (2004) examined group foraging behavior among humans by developing an experimental networked Java platform to create a common two- dimensional virtual world (an 80 x 80 grid) across computers. Participants sat at their respective computers and foraged for resources in real time by using the computers’ arrow keys to move up, down, left, and right in order to step on a food pellet and thereby consume it. We will briefly describe the experimental manipulations because their foraging environment and data serve as the initial basis for our agent model. Participants engaged in 6 five-minute sessions, consisting of all combinations of two perceptual conditions and three resource distribution conditions, and all participants experienced the same conditions in a given session. In the “visible” perceptual condition, a participant could see himself or herself as a yellow dot in the virtual world, and other participants were visible as blue dots while available food pellets were represented as green dots. The visible condition is therefore a good match for the assumptions of IFD (Fretwell and Lucas, 1970). In the “invisible” perceptual condition, a participant could see himself or herself as a yellow dot in the virtual world, but no other participants or food were visible in the world. The invisible condition corresponds to foraging experiments with sampling under uncertainty. The visible and invisible conditions thus represent two ends of a foraging perceptual spectrum. A new food pellet was dropped in one of two resource pools every 4/N seconds (where N is the number of participants), and there were three distribution conditions that probabilistically determined which pool received the pellet: 50/50, 65/35, and 80/20. For example, in the 65/35 distribution condition, 65% of food arrived at one pool while 35% arrived at the other pool. At each pool, new pellets were dropped according to a Gaussian distribution with a mean at the pool’s center and a variance of 5 units horizontally and vertically. Food release was constrained so that only one pellet could occupy a cell at a given time, and resource pool locations changed from session to session. In the invisible perceptual condition, a pellet appeared on the screen for two seconds for the participant who stepped on it, so participants could gradually ascertain the locations of the resource pools by exploring the world and occasionally obtaining pellets.