A fuzzy framework for modeling multiagent societies

摘要

In the past several years, the work within the framework of the interactivist-expectative theory on agency and learning (IETAL), we have concentrated on exploring the concept of learning environment partitions in an autonomous agent, and the problems encountered during its stay in the environment. The key concepts in the uniagent version of the theory are the concepts of expectancy and learning through interactions with the environment, while building an intrinsic model of it. Depending on the set of active drives and their hierarchy (not necessarily ordered with a partial ordering, but rather a general relational structure), the agent uses its intrinsic model to navigate its quest to satisfy the active drive(s). In this paper, we start with the existing results of the theory and generalize it to a theory of multiple agent systems, via introducing imitation-based interaction between homogenous agents. While sensing each other, the agents exchange their contingency tables, built during the interaction with the environment. This approach is inspired by results from both neurophysiology and psychology on the phenomenon of imitation. The formalization of the agent, as well as of the environment is necessary for setting up a successful experimental and simulation environment for exploring the new paradigms. Our approach relies on new modeling strategies and structures from the domain of lattice (L-fuzzy), posets (P-fuzzy) and general relational-structure-valued (R-fuzzy) algebraic structures.