The Iterated Prisoner's Dilemma: early experiences with Learning Classifier System-based simple agents

Prior research on artificial agents/agencies involves entities using specif ically tailored operational strategies (e.g., for information retrieval, pu rchase negotiation). In some situations, however, an agent must interact wi th others whose strategies are initially unknown and whose interests may co unter its own. In such circumstances, pre-defining effective counter-strate gies could become difficult or impractical. One solution, which may be viab le in certain contexts, is to create agents that self-evolve increasingly e ffective strategies from rudimentary beginnings, during actual deployment. Using the Iterated Prisoner's Dilemma (IPD) problem as a generic agent-inte raction setting, we use the Learning Classifier System (LCS) paradigm to co nstruct autonomously adapting "simple" agents. A simple agent attempts to c ope by maintaining an evolving but potentially perennially incomplete and i mperfect knowledge base. These agents operate against specifically tailored (non-adaptive) agents. We present a preliminary suite of simulation experi ments and results. The promise evidenced leads us to articulate several add itional areas of interesting investigations that we are pursuing. (C) 2001 Elsevier Science B.V. All rights reserved.