AGENTIC AI AND REINFORCEMENT LEARNING: TOWARDS MORE AUTONOMOUS AND ADAPTIVE AI SYSTEMS
DOI:
https://doi.org/10.47750/Keywords:
Agentic AI, Autonomous Systems, Adaptive AI, Reinforcement Learning, AI Trajectory, Utility Functions, Evolutionary Requirements, Agentic Capabilities, Axiomatic Values, Dynamic Environments, AI Design Philosophy, AI Research, AI Adaptability, AI Autonomy, Software Prototypes, Use Cases, AI Integration, AI Ethics, AI Evolution, Intelligent Agents.Abstract
This essay presents agentic AI, its functioning in a complex dynamic environment, and the need
for developing AI systems that are more autonomous and adaptive, working as an agent. The
conclusion of requirements for agentic AI, in terms of agentic capabilities and axiomatic values,
leads to the basic design philosophy of constructing AI systems that approximate such agentic
perception and axioms, rather than trying to explicitly describe or entangle human meta-morals
into the AI systems. Thus, we propose to focus on designing more autonomous and adaptive AI
systems that are more similar to agentic entities whose utility functions are developed from the
evolutionary functional requirements. Reinforcement learning methods permeate mainstream AI
research since they often provide the best empirical results. Due to that, we focus our exploration
on research into integrating agentic capabilities with RL methodologies. On the basis of software
prototypes, we exemplify the problem of needing to wait for a print job and an RL-based
solution for a delivery service. Further exploration into the research goals and full conclusions
are covered in the debriefing of each use case.
