清华新研究，Nature+Science双杀

Core Insights - The research from Tsinghua University reveals a paradox in the AI for Science field, where AI accelerates individual scientific progress but narrows collective scientific focus, leading to a phenomenon known as "group climbing" [2][4][17] - Despite AI's contributions to scientific advancements, such as the development of AlphaFold, the overall disruptive research outcomes across disciplines have been declining [4][17] Group 1: Research Background - The study aims to address the contradiction of why overall scientific progress has not significantly accelerated despite the empowerment of research by AI [4] - The research culminated in the publication of the paper titled "Artificial Intelligence Tools Expand Scientists' Impact but Contract Science's Focus" [4][5] Group 2: Methodology - The research team utilized a combination of expert annotation and large-scale language model inference to identify AI-augmented research from a dataset of 41.3 million research papers [5][6] - The methodology involved creating a "semantic map" of scientific literature to analyze the impact of AI across six major fields of natural sciences [9][10] Group 3: Findings on Individual vs. Collective Impact - Individual scientists using AI publish 3.02 times more papers and receive 4.84 times more citations, becoming project leaders 1.37 years earlier than their non-AI counterparts [15][16] - However, at the collective level, the knowledge breadth of AI-augmented research decreased by 4.63%, and cross-disciplinary interactions among scientists fell by 22% [15][16] Group 4: Underlying Causes - The decline in collective scientific breadth is attributed to the lack of generalizability in current AI models, which leads researchers to focus on a limited number of popular research areas suitable for AI [17][20] - This "group climbing" effect accelerates solutions to known problems but restricts exploration of unknown areas, resulting in a trade-off where breadth is sacrificed for speed [17][20] Group 5: Future Directions - To address these limitations, the research team introduced the OmniScientist system, designed to provide comprehensive, cross-disciplinary support for scientific research, evolving AI from a mere tool to an "AI scientist" capable of hypothesis generation and experimental design [20]