Assessing Large Language Models for Decision Support in Novel Circular Supply Chains: A Case of Hotel Linen
研究比较了专家与多个大语言模型在酒店布草回收的障碍识别和关系评估任务中的表现，发现LLM在依赖通用知识的任务上可替代专家，但在需要具体情境知识的任务上表现不足。
Decision tasks involved in implementing novel circular supply chain (CSC) modes entail significant challenges due to inherent contextual uncertainty and extensive information requirements. This paper investigates the potential of large language models (LLMs) to support such knowledge intensive decision tasks. Focusing on the barrier prioritization task for used hotel linen recycling, we compare the performance of an expert panel and several LLMs on two subtasks: barrier identification (a knowledge-level task) and barrier interrelationship assessment (a wisdom-level task). To ensure research rigor and generalizability, we conducted extensive robustness checks across multiple state of-the-art LLMs (including GPT-5, Gemini-2.5-Pro, and Qwen-Max). Results indicate that LLMs underperform in identifying context-specific barriers but match experts in assessing inter-barrier relationships once domain knowledge is provided. These findings suggest that “task generality” should serve as a critical boundary for assigning responsibilities between human experts and LLMs in CSC implementation: tasks reliant on general or pre-supplied domain knowledge may be delegated to LLMs, whereas tasks tightly coupled with specific operational contexts necessitate the tacit knowledge and direct experience of human experts. Our research contributes to the literature on human-AI collaboration, supply chain management, and Organizational Information Processing Theory.