A Theoretical Framework for AGI Based on the Synergistic Operation of Multiple Modules-From human,To the infinite
Year: 2025
Type: dissertation
Abstract: In recent years, the emergence of large language models (LLMs) has greatly advanced AGI research, revealing their potential for tasks ranging from natural language understanding to complex reasoning. This paper presents a theoretical framework for LLM-based knowledge-injection neural networks, achieving comprehensive AGI functionality through the collaborative interaction of seven modules: the Perception Module (based on few-shot learning and change-driven sensory entropy), the Logical Reasoning Module (supporting natural language logical mapping and dynamic decision-making), the Autonomous Consciousness Module (based on Bayesian generalization and behavioral reflection), the Metric Module (dynamically assessing commonalities and differences), the Memory/Knowledge Module (storing unique memories and experiences while integrating LLM with internet-based learning), the Emotion Module (constructing emotions and values through hormone-driven mechanisms), and the Physical Environment Simulation Module (modeling environmental sequences and conducting virtual experiments). Bidirectional LLM mapping facilitates communication between modules, and all modules incorporate internet-based self-learning to handle new tasks. Key innovations include natural language-based logical reasoning, hormone-driven desire mechanisms, dynamic metric functions for common feature extraction, change-driven perception prioritization, consequence prediction through sensory sequences, cognitive leaps triggered by free will, adaptive self-learning through the internet, and modular design prospects. Through detailed theoretical elaboration and interdisciplinary perspectives, this work provides a comprehensive foundation for constructing a complete AGI entity.
Author: Yan, Kailin
Institution: China University of Petroleum, Beijing
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W6911972886
Cites:
Cited by:
Related to: