Large language models (LLMs) demonstrate impressive performance on a wide variety of tasks, but they often struggle with tasks that require multi-step reasoning or goal-directed planning. To address this, we take inspiration from the human brain, in which planning is accomplished via the recurrent interaction of specialized modules in the prefrontal cortex (PFC). These modules perform functions such as conflict monitoring, state prediction, state evaluation, task decomposition, and task coordination. We find that LLMs are sometimes capable of carrying out these functions in isolation, but struggle to autonomously coordinate them in the service of a goal. Therefore, we propose a black box architecture with multiple LLM-based (GPT-4) modules. The architecture improves planning through the interaction of specialized PFC-inspired modules that break down a larger problem into multiple brief automated calls to the LLM. We evaluate the combined architecture on three challenging planning tasks -- graph traversal, Tower of Hanoi, and logistics -- finding that it yields significant improvements over standard LLM methods (e.g., zero-shot prompting, in-context learning, and chain-of-thought). These results demonstrate the benefit of utilizing knowledge from cognitive neuroscience to improve planning in LLMs.

翻译：大型语言模型（LLM）在各类任务中表现出令人瞩目的性能，但在需要多步推理或目标导向规划的任务中往往表现不佳。为解决这一问题，我们借鉴人脑机制——其规划功能通过前额叶皮层（PFC）中专化模块的循环交互实现。这些模块分别执行冲突监控、状态预测、状态评估、任务分解及任务协调等职能。我们发现LLM虽能独立执行部分功能，却难以自主协调这些功能以服务整体目标。为此，我们提出一种基于多LLM（GPT-4）模块的黑盒架构。该架构通过模拟前额叶皮层的专化模块交互来增强规划能力，将复杂问题分解为多次简短的LLM自动调用。我们在三项具有挑战性的规划任务（图遍历、汉诺塔、物流规划）上评估该组合架构，发现其相较于标准LLM方法（如零样本提示、上下文学习、思维链）取得了显著性能提升。这些结果证明，利用认知神经科学知识来改进LLM规划能力具有重要价值。