The El Farol Bar Problem and Computational Effort: Why People Fail to Use Bars Efficiently

Does how much an agent thinks about its own actions affect the global properties of a system? We use the El Farol Bar Problem to investigate this question. In this model, the El Farol Bar represents a scarce resource. Does the amount of computational ability that agents possess affect resource utilization? For instance, if agents attend the bar randomly on average 50 people will go to the bar. On the other hand, if agents act as neoclassical economics suggest, its not clear what the average attendance at the bar will be, but in this paper we will argue that it will also be near 50. In Arthur’s original model, he showed, using a simulation involving an ecology of strategies, that the average attendance of the bar converged to 60. Fogel et al. gave their agents more computational power and let them use a evolutionary algorithm; they showed that the average attendance at the bar was 56, not 60. If we examine these four results of (1) random agents, (2) perfect agents, (3) Arthur’s agents, and (4) Fogel et al.’s agents, we can ask whether there is a relationship between computational effort and attendance at the bar (e.g. the utilization of a public resource). To investigate this question we look at a model where we can control the computational power that each agent has to predict the attendance each week.