Emergent Temporal Awareness in Reinforcement Learning Agents through NSI Dynamics

This article presents the Negentropic Stabilization Index (NSI) as an internal temporal metric in reinforcement learning (RL) agents. NSI, defined as the negative gradient of entropy over a sliding window of states, enables agents to internally segment time into phases such as exploration and consolidation. Through a CartPole environment, we demonstrate how NSI dynamics correspond to shifts in learning behavior, suggesting a pathway toward emergent temporal awareness in artificial systems.