Modular Cognitive Architecture Emerges in Large Language Models

大型语言模型中涌现出模块化认知架构

Abstract: The human brain exhibits a striking degree of functional specialization, with distinct networks supporting language, formal reasoning, reasoning about other minds, and reasoning about the physical world. Is this modular organization a fundamental principle of how intelligent systems must be built, or an evolutionary accident specific to biological brains?

摘要： 人类大脑表现出惊人的功能专业化程度，拥有支持语言、形式推理、心智理论（对他人的思维进行推理）以及物理世界推理的独特网络。这种模块化组织是构建智能系统的基本原则，还是生物大脑特有的进化偶然？

Here, we test whether a similar organization emerges in Large Language Models—another class of intelligent systems created through a very different optimization process. Using circuit analyses across N=46 tasks spanning four cognitive domains (language, formal reasoning, social reasoning, physical reasoning), we find that LLMs develop a modular architecture that mirrors the human brain: tasks drawing on the same network in humans recruit overlapping neurons in LLMs, whereas tasks drawing on different networks recruit distinct neurons.

在此，我们测试了在大型语言模型（LLM）——这一通过截然不同的优化过程创建的智能系统——中是否也会涌现出类似的组织结构。通过对涵盖四个认知领域（语言、形式推理、社会推理、物理推理）的 46 项任务进行电路分析，我们发现 LLM 发展出了一种与人类大脑相呼应的模块化架构：在人类中调用相同网络的任务，在 LLM 中会招募重叠的神经元；而调用不同网络的任务，则会招募不同的神经元。

The convergent emergence of modularity in brains and neural networks suggests that it may be a fundamental property of intelligent systems.

大脑和神经网络中模块化特征的趋同涌现表明，这可能是智能系统的一种基本属性。