Belief convergence to facilitate cooperative behaviors

In Artificial Intelligence (AI), utility functions are used to compare the relative goodness of an AI system making one decision over another. These utility functions, along with their coefficients and parameters, comprise a set of beliefs. The term “belief” is used to describe how autonomous systems associate quantified confidences in the existence of things that make up their worldly knowledge; hence, decisions made by an AI system are governed by the states of its belief. Just as in control system theory where the states are quantities used to estimate and describe the behaviors of electrical and/or mechanical systems, belief states are quantities that are used to describe and estimate decision making characteristics. For evolving AI systems (e.g., ones that change their belief states over time through learning) and heterogeneous systems (e.g., ones that do not share a common notion of a global belief), it is important to develop mechanisms that will allow these systems to converge their beliefs, where they can determine how each other “thinks.” This is important since it will allow each agent in a cooperating collective to make decisions that optimally solves their individual needs, along with their collective needs, without requiring explicit communication or a mediator. In this paper, methodologies are presented that facilitate belief convergence. The significance of this study is that we demonstrate how a state observer, such as a Kalman filter, can be used by each agent in a collective to estimate neighboring belief states. This is done with no a priori information of their neighbors' belief states, and by only comparing each individual's estimate for the required effort of the whole collective to perform a group plan.