Gödel Agent: A Self-Referential Agent Framework for Recursive Self-Improvement

Xunjian Yin,Xinyi Wang,Liangming Pan,Xiaojun Wan,William Yang Wang

2024-10-18

Abstract:The rapid advancement of large language models (LLMs) has significantly enhanced the capabilities of AI-driven agents across various tasks. However, existing agentic systems, whether based on fixed pipeline algorithms or pre-defined meta-learning frameworks, cannot search the whole agent design space due to the restriction of human-designed components, and thus might miss the globally optimal agent design. In this paper, we introduce Gödel Agent, a self-evolving framework inspired by the Gödel machine, enabling agents to recursively improve themselves without relying on predefined routines or fixed optimization algorithms. Gödel Agent leverages LLMs to dynamically modify its own logic and behavior, guided solely by high-level objectives through prompting. Experimental results on mathematical reasoning and complex agent tasks demonstrate that implementation of Gödel Agent can achieve continuous self-improvement, surpassing manually crafted agents in performance, efficiency, and generalizability.

Artificial Intelligence

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to design an agent framework that can be autonomously and recursively improved, in order to overcome the limitations of existing agent systems in design space exploration. Existing agent systems, whether based on fixed algorithmic processes or predefined meta - learning frameworks, are limited by human - designed components and are unable to search the entire agent design space, and thus may miss the globally optimal design solution. The paper proposes a self - referential framework named Gödel Agent, which is inspired by the Gödel machine. This framework enables agents to dynamically modify their own logic and behavior through large - language models (LLMs) without relying on preset programs or fixed optimization algorithms, and to achieve self - recursive improvement only through high - level goal guidance. Experimental results show that Gödel Agent can achieve continuous self - improvement in the implementation of multiple fields such as programming, science, and mathematics, and its performance, efficiency, and generalization ability exceed those of manually - designed agents.