A Study on Foundation Language Models and the Role of Episodic Memory in Agentic Artificial Intelligence (Agentic AI)
A Study on Foundation Language Models and the Role of Episodic Memory in Agentic Artificial Intelligence (Agentic AI)
Hamid Mojarad1 Mohammad Hasan Nasehi2 Seyedeh Sali Samiei3
1) Ph.D. Candidate in Artificial Intelligence, Islamic Azad University, Kashan Branch, Iran.
2) Ph.D. Candidate in Artificial Intelligence, Islamic Azad University, Kashan Branch, Iran.
3) M.Sc. Student in Software Engineering, Islamic Azad University, Tehran Branch, Iran.
IOI Code
:
AAHD-AAKFKG
Publication :
National Conference on Information Technology, Nanotechnology, Artificial Intelligence and Technological Futures Studies(itnaf.ir)
Abstract :
In recent years, Large Language Models (LLMs) have emerged as the core of generative artificial intelligence, driving a remarkable transformation in the understanding and generation of natural language. However, due to their static structure and lack of persistent memory, these models face challenges in managing long-term interactions and recalling past experiences. The aim of this study is to conceptually and analytically investigate the role of episodic memory in enhancing the performance of large language models within agentic environments. Using a review-based research method, scientific papers published between 2022 and 2025 were examined to analyze the mechanisms of integrating memory with transformer architectures. The findings indicate that incorporating episodic memory modules into foundation models enhances reasoning ability, contextual coherence in dialogue, and decision-making performance in intelligent agents. Moreover, the adoption of agentic frameworks such as ReAct and MemGPT has paved new paths toward developing artificial intelligence systems with cognitive continuity and human-like behaviors. Ultimately, the results suggest that integrating foundation models with episodic memory can lead to the emergence of intelligent agents that, instead of merely reacting, demonstrate cognitive persistence, adaptive behavior, and experiential learning capabilities.
Keywords : large language models episodic memory agentic artificial intelligence transformer foundation models deep learning