A survey of decision-making and action selection methods in artificial general intelligence

A survey of decision-making and action selection methods in artificial general intelligence

Mohammad Moeid Jamalimehr1 Hossain Saberi2

1) PhD student in Computer Science and Engineering, Artificial Intelligence, Imam Hussein University (IHU), Iran -
2) Assistant Professor, Imam Hossein University (IHU), Iran -

IOI Code : AAHD-ABKHBK
Publication : National Conference on Information Technology, Nanotechnology, Artificial Intelligence and Technological Futures Studies(itnaf.ir)
Abstract :
    Artificial General Intelligence (AGI) which aims to imitate human cognitive capabilities, is considered one of the fundamental challenges in modern AI. In this context, action selection, as a core cognitive ability, plays a critical role in how intelligent agents interact with dynamic environments. In this survey, we investigate various methodologies and techniques for developing Action Selection Mechanisms (ASM). We also propose a novel taxonomy, categorizing these approaches into knowledge-based techniques (e.g., temporal planners and task decomposition), learning-based techniques (e.g., instance-based learning and reinforcement learning), and attention-based techniques (with an emphasis on cognitive information reduction and Large Language Model-based game agents). Ultimately, this study demonstrates that despite the abundance of techniques designed to enhance decision-making systems, no single technique is sufficient to able agents in complex environments. Consequently, the integration of techniques in these three main categories into a unified framework is a necessary step for the development of autonomous agents and the eventual realization of Artificial General Intelligence.

Keywords : artificial general intelligence decision making action selection reinforcement learning large language models