Beyond the Bayesian Truth Serum: The Knowledge Free Peer Prediction Mechanism

The elicitation of private information from individuals is crucially important to many tasks, ranging from scientific research to corporate decision-making. Eliciting private information is particularly challenging when objective truth is inaccessible when there is no “anwer key” available. To address this challenge, we present the Knowledge Free Peer Prediction mechanism (KFPP). KFPP induces truthful reporting for any number of agents n ≥ 3, doesn’t require the mechanism to know the common prior, and can handle non-binary information elicitation; it thus improves on previous information elicitation mechanisms designed for this setting, like Peer Prediction, the Bayesian Truth Serum, and the Robust Bayesian Truth Serum. Furthermore, we demonstrate that KFPP can handle several complications, including risk-adverse participants, continuous signals, and participants who experience varying costs when acquiring and reporting their information.