Comparing AI and human decision-making mechanisms in daily collaborative experiments
Linghao Wang,
Zheyuan Jiang,
Chenke Hu,
Jun Zhao,
Zheng Zhu,
Xiqun Chen,
Ziyi Wang,
Tianming Liu,
Guibing He,
Yafeng Yin,
Der-Horng Lee
Affiliations
Linghao Wang
Institute of Intelligent Transportation Systems, College of Civil Engineering and Architecture, Zhejiang University, Hangzhou, China
Zheyuan Jiang
Institute of Intelligent Transportation Systems & Polytechnic Institute, College of Civil Engineering and Architecture, Zhejiang University, Hangzhou, China
Chenke Hu
Institute of Intelligent Transportation Systems & Polytechnic Institute, College of Civil Engineering and Architecture, Zhejiang University, Hangzhou, China
Jun Zhao
Department of Civil and Architectural Engineering and Mechanics, The University of Arizona, Tucson, AZ, USA
Zheng Zhu
Institute of Intelligent Transportation Systems, College of Civil Engineering and Architecture, Zhejiang University, Hangzhou, China; Corresponding author
Xiqun Chen
Institute of Intelligent Transportation Systems, College of Civil Engineering and Architecture, Zhejiang University, Hangzhou, China; Corresponding author
Ziyi Wang
Zhejiang Tranalytic Technology Company, Wenzhou, China
Tianming Liu
Department of Civil and Environmental Engineering, University of Michigan, Ann Arbor, MI, USA
Guibing He
Department of Psychology and Behavioral Sciences, Zhejiang University, Hangzhou, China
Yafeng Yin
Department of Civil and Environmental Engineering, University of Michigan, Ann Arbor, MI, USA
Der-Horng Lee
Zhejiang University–University of Illinois Urbana Champaign Institute, Zhejiang University, Hangzhou, China
Summary: Artificial intelligence (AI) is trying to catch up with human beings in many aspects. In this track, the potential for replacing human decision-making with AI models, such as large language models (LLMs), has become a topic of considerable debate. To test the performance of AI in daily decision-making, we compared humans, LLMs, and reinforcement learning (RL) in a multi-day commute decision-making game. It denotes a collaborative decision-making process where individual and collective outcomes are interdependent. We examined various performance metrics, including overall system results, system convergence progress, individual decision dynamics, and individual decision mechanisms. We find that LLMs exhibit human-like abilities to learn from historical experience and achieve convergence when making daily commute decisions. However, in the context of multi-person collaboration, LLMs still face challenges, such as weak perception of others’ choices, poor group decision-making mechanisms, and a lack of physical knowledge.
