Title: The Interplay of Cognitive Diversity: A Simulation Study of Behavioral Biases in Multi-Tier Supply Chains Using LLMs

Abstract:

Achieving effective coordination among generative agents within complex, multi-round decision-making frameworks remains a significant hurdle for both artificial intelligence and operations management. While behavioral studies have previously identified cognitive biases as root causes of supply chain inefficiencies, conventional experimental approaches are often constrained by issues of scalability and control. To address these limitations, this research presents a scalable experimental framework that leverages Large Language Models (LLMs) to mimic multi-stage supply chain dynamics. Anchored in a Hierarchical Reasoning Framework, our investigation focuses on how cognitive heterogeneity shapes interactions among agents. Departing from previous studies that relied on homogeneous setups, we utilize DeepSeek and GPT agents to deliberately vary the sophistication of reasoning capabilities across different tiers of the supply chain. Through simulations that are both rigorously replicated and statistically validated, we examine the influence of this cognitive diversity on collective performance. Our data reveals that agents tend to display self-interested and myopic behaviors, which ultimately deepen systemic inefficiencies. Nevertheless, the study demonstrates that the implementation of information sharing mechanisms can effectively counteract these negative outcomes. These results not only expand upon traditional behavioral methodologies but also provide fresh perspectives on the operational dynamics of AI-integrated organizations. Ultimately, this work highlights both the capabilities and the constraints of employing LLM-based agents as stand-ins for human decision-makers in intricate operational contexts.

Source: arXiv Generated at: 2026-06-03 00:00:00 UTC