Date of Award

12-1-2025

Degree Name

Master of Science

Department

Computer Science

First Advisor

Hexmoor, Henry

Abstract

This case study presents a novel cognitive architecture that integrates Global Workspace Theory (GWT) and Predictive Processing (PP) to model synthetic consciousness in a reinforcement learning (RL) agent. Implemented within the MiniGrid environment, the proposed agent learns to navigate and perform goal-directed tasks under partial observability using minimal supervision. The system incorporates specialized modules for sensation, prediction, emotion, memory, and active policy, unified through a global workspace that enables attentional broadcasting and adaptive decision-making. Learning occurs through continuous prediction-error minimization and free-energy reduction, allowing the agent to form internal representations, update beliefs, and maintain homeostatic balance across drives such as energy, threat, and curiosity. Empirical results demonstrate that the agent exhibits emergent cognitive dynamics including stable prediction convergence, periodic workspace activation, and emotionally modulated behavior, indicative of adaptive and self-regulating intelligence. This work contributes a computational framework for exploring conscious-like learning mechanisms in artificial agents and provides insights into the intersection of cognitive neuroscience and machine learning.

Share

COinS
 

Access

This thesis is only available for download to the SIUC community. Current SIUC affiliates may also access this paper off campus by searching Dissertations & Theses @ Southern Illinois University Carbondale from ProQuest. Others should contact the interlibrary loan department of your local library or contact ProQuest's Dissertation Express service.