_
Abstract
This project develops a framework for computational general intelligence that can be used to build computational models of decision making for reinforcement learning agents that operate in the real world. The framework proposes a novel definition of information and proposes virtual neurons as a method to enable longer term prediction and planning for agent decision making. Virtual neurons function as the most fundamental building blocks of an agent's cognitive model of the world and are the mediating particles of an agent's interaction with the proposed information field, and represent the set of an agent's available receptive fields that span multiple layers of abstraction. Layers of virtual neurons are computationally represented by graph structures to enable path planning, task navigation, and enable decision making and coordination with other agents through the creation of a mutual goal space.