
Abstract: By taking the “Manhattan Project” as a case study, this paper explores the application of large language models (LLMs) for counterfactual reasoning in the history of science and technology. Through multiple rounds of experiments using DeepSeek-V3.2, ERNIE Bot 5.0 Preview, and GLM-4.6, a strong consensus has been observed: the absence of “Attack on Pearl Harbor” would have led to a delayed initiation of the Manhattan Project, reduced resource investment, and a shift from a military-centralized to an academically decentralized organization. The findings demonstrate that LLMs can effectively simulate alternative historical pathways, revealing the profound dependence of technological development on political context and social mobilization, thus offering a systematic and verifiable new approach for research of history of science and technology.
Key Words: Large language models; Counterfactual reasoning; History of science and technology; Manhattan Project; STS
