Trust Revelation in Multiagent Interaction

We analyze untrustworthy interactions, that is, interactions in which a party may fail to carry out its obligations. Such interactions pose agents with the problem of how to estimate the trustworthiness of the other party. The efficiency of untrustworthy interactions critically depends on the amount and the nature of information about untrustworthy agents. We propose a solution to the problem of learning and estimating trustworthiness. Instead of relying on a third party for providing information or for backing up multiagent interaction, we propose an incentivecompatible interaction mechanism in which agents truthfully reveal their trustworthiness at the beginning of every interaction. In such a mechanism agents always report their true level of trustworthiness, even if they are untrustworthy.