The extent to which Artificial Intelligence (AI) can trigger generalized paradigm shifts in science is unclear. Although some of these technologies have revolutionized data collection and analysis in specific scientific fields such as Chemistry, their overall impact depends on the scope of adoption and the ways scholars use them. In this study, we document substantial differences in the timing and extent of AI adoption across countries and scientific domains from 1960 to 2015. After 2015, we find generalized exponential growth in AI adoption, with the number of AI-supported works multiplying by at least four across all domains. The transformative nature of this rapid growth is less apparent and points to multiple challenges should adoption trends persist. According to our analyses, AI-supported research is confined to very few topics with strong ties to Computer Science and conventional statistical frameworks, suggesting limited transformational potential in epistemological terms. AI-supported works are also associated with an unwarranted citation premium and exhibit substantially higher retraction rates than non-AI-supported works across most fields. Geographically, AI adoption displays pronounced heterogeneity at the country level, along with an acceleration in the relevance of middle-income countries in Asia, from China and beyond. Thus, the transformative capacity of AI in science remains largely untapped, and its rapid adoption underlines challenges in research openness, transparency, reproducibility, and ethics from a global perspective. We discuss how best research practices could boost the benefits of AI adoption and highlight fields and geographies where these trends warrant closer scrutiny.

翻译：人工智能能否引发科学领域的普遍范式转变尚不明确。尽管某些AI技术（如化学领域）已彻底改变了特定学科的数据收集与分析方式，但其整体影响取决于应用的广度及学者使用方式。本研究系统记录了1960年至2015年间，不同国家与科学领域在AI应用时间点及覆盖范围上的显著差异。2015年后，AI应用呈普遍指数级增长，所有领域内AI辅助研究的数量至少翻了两番。然而，这种高速增长的变革性特征并不明显：若应用趋势持续，或将引发多重挑战。分析表明，AI辅助研究集中在与计算机科学及传统统计框架紧密关联的极少数主题，在认识论层面展现出有限的变革潜力。此外，AI辅助研究存在不合理的引文优势，且多数领域的撤稿率显著高于非AI辅助研究。地理分布上，AI应用在国家层面呈现显著异质性，同时以中国为代表的亚洲中等收入国家的相关性加速提升。由此可见，AI在科学中的变革能力尚未充分释放，其快速普及凸显了全球视角下研究开放性、透明度、可重复性及伦理面临的挑战。本文探讨了如何通过优化研究实践来提升AI应用的效益，并指出需要重点关注的学科领域与地理区域。