AI strategy under institutional pressure: strategic conformity and decision-making in large language models
通过2400条AI生成战略建议与218名管理者的对比实验，发现大语言模型在制度压力下系统性偏好战略顺从，缺乏人类决策者的创造性抵抗，并引入“合成乐观”概念解释这一偏差。
• 2,400 LLM generated strategic recommendations compared against a human benchmark of 218 managers using Oliver’s typology. • LLMs systematically favor strategic acquiescence under institutional pressure relative to human decision makers. • Introduces “Synthetic Optimism” as a conformity driven bias in AI supported strategic decision-making. • LLMs act as institutional amplifiers, reinforcing norms and compressing strategic response diversity. Organizations increasingly integrate artificial intelligence into corporate strategy to gain a competitive edge, yet these tools may paradoxically drive strategic homogenization. We examine how large language models respond to institutional pressures using a computational experiment generating 2400 strategic recommendations across ten business cases and four leading models, complemented by a human benchmark study of 218 managers. Our results expose a strategic paradox: AI lacks strategic agency and functions instead as an institutional amplifier that favors compliance over the creative resistance displayed by human decision-makers. We explain this divergence by introducing the concept of Synthetic Optimism, a structural bias of AI toward hyper conformity and non-resistant alignment with institutional consensus that yields a predictable conformity baseline. For organizations using LLMs in strategic decision-making, we translate this into actionable levers to surface industry norms, test deliberate divergence, and identify strategic blind spots: Institutional Mapping, Strategic Resistance Toggling, and Institutional Tension Simulation.