Modeling minority games with BDI agents – a case study

  • Authors:
  • Wolfgang Renz;Jan Sudeikat

  • Affiliations:
  • Multimedia Systems Laboratory, Hamburg University of Applied Sciences, Hamburg, Germany;Multimedia Systems Laboratory, Hamburg University of Applied Sciences, Hamburg, Germany

  • Venue:
  • MATES'05 Proceedings of the Third German conference on Multiagent System Technologies
  • Year:
  • 2005

Quantified Score

Hi-index 0.00

Visualization

Abstract

Binary decisions are common in our daily lives and often individuals can gain by choosing the minority’s side. The socio–economically inspired Minority Game (MG) has been introduced as an exact model of the famous El Farol’s Bar Problem, which exhibits complex behavior. In this paper we show that the MG players can be naturally modeled by agents using reactive planning, implemented with a common deliberative programming paradigm, the Belief–Desire–Intention (BDI) model. Our simulation framework is build in Jadex, a forthcoming platform implementing BDI notions. Straightforward implementation of multi–agent simulations is enabled by XML agent descriptions and referenced Java classes. Design of the player agents and simulation results are shown. As a case study, we introduce a new adaptive stochastic MG with dynamically evolving strategies. It exhibits different regimes, reaching from optimal cooperation to destructive behavior, including the emergence of the so called ”Schwarzer Peter” game, depending on control parameters. We identify optimization mechanisms like rotation in the working regime as well as metastable behavior.