Active Bayesian observer correcting overconfidence effects due to E-type confirmatory bias

This paper presents a preliminary analysis of an active Bayesian observer that communicates with humans to correct the overconfidence effects due to their confirmatory biases. Two types of confirmatory biases, called Z- and E-types, are introduced and formalized using a model of noisy channels between signals and perceptions. Persons suffering from Z-type confirmatory bias receive a signal and perceive it incorrectly as a different signal with a certain probability, while those suffering from E-type confirmatory bias fail to receive a signal with a certain probability. A dynamic Bayesian network model is developed to analyze the effects of Z- and E-types in a unified way. The analytical model enables us to give a theoretical insight into some basic properties of the active Bayesian observers.