A Model for Competence and Integrity in Variable Payoff Games

Agents often have to trust one another when engaging in joint actions. In many cases, no single design team has the authority to assure that agents cooperate. Trust is required when agents hold potentially different values or conflicting goals. This paper presents a framework and some initial experiments for decomposing agent reputation within a multi-agent society into two characteristics: competence and integrity. The framework models competence as the probability of successfully carrying out an intended action. Integrity is modeled as a rational commitment to maintaining a reputation, based on the agent’s assessment of the game’s discount rate. We show that a simple, one-level-deep recursive model—given accurate knowledge of self and the other agent’s competence and integrity (commitment to reputation)—outperforms titfor-tat and other standard strategies in evolutionary round-robin iterated prisoner’s dilemma tournaments. This indicates that the approach taken here warrants further investigation using more realistic and com-