语言模型的认知地图：通过口头表达世界模型实现最优规划

Cognitive Map for Language Models: Optimal Planning via Verbally Representing the World Model

June 21, 2024

作者: Doyoung Kim, Jongwon Lee, Jinho Park, Minjoon Seo

cs.AI

摘要

语言模型在各种自然语言处理任务中展现了令人印象深刻的能力，但在需要多步模拟的规划任务中却遇到了困难。受人类认知过程启发，本文研究了能够构建给定环境认知地图的语言模型的最佳规划能力。我们的实验表明，认知地图显著提升了格子世界路径规划任务中最佳和可达规划生成能力的性能。我们观察到我们的方法展示了与人类认知相似的两个关键特征：将其规划能力泛化到外推环境以及在有限训练数据下快速适应。我们希望我们在格子世界任务中的发现能够为在语言模型中建模人类认知过程提供见解，潜在地促进更先进和更健壮系统的发展，使其更好地类似于人类认知。

English

Language models have demonstrated impressive capabilities across various natural language processing tasks, yet they struggle with planning tasks requiring multi-step simulations. Inspired by human cognitive processes, this paper investigates the optimal planning power of language models that can construct a cognitive map of a given environment. Our experiments demonstrate that cognitive map significantly enhances the performance of both optimal and reachable planning generation ability in the Gridworld path planning task. We observe that our method showcases two key characteristics similar to human cognition: generalization of its planning ability to extrapolated environments and rapid adaptation with limited training data. We hope our findings in the Gridworld task provide insights into modeling human cognitive processes in language models, potentially leading to the development of more advanced and robust systems that better resemble human cognition.