EconAI: Preference-driven Agents Simulating Economic Activities via Large Language Model

Zongsheng Cao, Anran Liu, Jun Xie, Feng Chen, Zhepeng Wang

Keywords: Knowledge Agent, Economic Simulating

Abstract: The emergence of artificial intelligence has transformed the methodological frameworks in economic research by simulating intricate interactions among diverse agents. Despite the advantage of large language models (LLMs), they often struggle with occasions involving decision-making interactions with environments. This challenge stems from the fact that most LLMs are rationality-driven, seeking optimal economic benefits, while humans are preference-driven, pursuing the balance of personal goals (\textit{e.g.,} income and health). These differences hinder the LLMs' ability to effectively understand economic activities across various contexts, leading to biases in economic simulations. To tackle this issue, we introduce \textbf{EconAI}, a novel approach aimed at enhancing the preference learning capabilities of LLMs by incorporating human-like preferences and cognitive processes. Specifically, EconAI features a 'knowledge brain' constructed from historical data and learning algorithms, enabling memory and making decisions for sophisticated economic facts. By integrating elements of self-learning, reflection, and experience updates, we refine decision-making processes, resulting in more accurate economic planning and mitigating planning bias in economic activities. Through the integration of real-time economic data and historical trends, EconAI offers a robust simulation platform that can adapt to market fluctuations and economic shocks. Our findings demonstrate that EconAI can model economic phenomena like inflation and employment with greater precision, showcase a notable ability to adjust to changing economic conditions, and surpass existing frameworks significantly.

Primary Area: other topics in machine learning (i.e., none of the above)

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 10235