Cogent: A Coherence-Driven Cognitive Agent Modeling and Simulation Framework

Abstract

Agent based modeling is an effective methodology for understanding complex cognitive systems through generative mechanisms that explain emergent behavior. However, most agent modeling languages use general-purpose imperative language constructs and lack high-level syntactic features necessary for cognitive modeling. In this thesis, we present Cogent, a cognitive coherence-driven agent modeling and simulation platform and demonstrate its use in ethical decision-making by autonomous systems. The underlying strategy is based on parallel constraint satisfaction in terms of a connectionist interactive activation model that implements the theory of cognitive coherence. Agents in the Cogent language are specified by a Domain-Specific Language, which provides the syntax and semantics for creating the decision-making model along with its coherence network. A model in Cogent expresses the constraints of a cognitive architecture through mechanisms such as explanation, deliberation, deduction, and analogy. The language provides the ability to model complex hierarchical network structures while allowing the visualization of the coherence network to gain cognitive affordance into explaining an agent's behavior and decisions. To illustrate the utility of Cogent, we explore a case study in machine ethics and demonstrate how deontological and consequentilist approaches to decision-making in philosophical ethics can be simulated using the Cogent.