Towards a method for bootstrapping agent trust

Prediction of the partnerpsilas future performance is essential for an agent who aims at establishing cooperation relationships. Mutual interactions could fail because of limited competence or because of a poor commitment to the goal; the latter one is due either to the agentpsilas desire to cheat or to its intention of punishing the partner for some prior injuries suffered from it. This paper proposes a method for bootstrapping a competence model for an agent; the model assumes a normal distribution of unaltered experiences. It makes use of the agentpsilas prior experiences with honest, cheating or revenge-seeking partners which limits the impact of insincere (deliberate low commitment behavior) over the final model.