AI4S势不可挡！Nature重磅报告：调研上万科学家，超4成自费也要用AI
Ranking
No observed public metrics; popularity remains neutral/archived.
Merged summary
TL;DR - Fudan University, the Shanghai Academy of AI for Science, and Springer Nature surveyed 10,480 researchers across 117 countries and 12 disciplines on AI use in research — the largest such study to date — finding AI is now embedded in daily scientific work but blocked from trusted, accountability-bearing tasks by a persistent credibility gap.
- Usage splits by task type: high adoption in text/information work (manuscript polishing 44.6%, literature search 43.8%, research planning 40.9%) versus low adoption where accountability matters (peer review 23.5%, submission decisions 25.7%).
- General-purpose LLMs dominate at 75.9% of tool mentions, with specialized literature/data/visualization tools under 10% combined; ChatGPT leads globally at 36.8% mention rate, while DeepSeek reaches 48.8% among Chinese researchers but under 10% abroad.
- Institutional funding is scarce: only 11.1% get AI via institutional procurement, 45.4% use free tiers, and 41.3% pay out of pocket — highest self-funding among hospital staff (49.1%) — creating compliance blind spots from unmonitored personal accounts.
- Open-ended responses show concern outweighing enthusiasm: accuracy/hallucination drew 4,935 mentions versus 3,442 for productivity gains, followed by weakened critical thinking (1,167), integrity damage from fabricated content (702), and data/IP leakage (216) — a paradox given that citation traceability is the top cited benefit.
- Generational divide: 46.8% of researchers with under 3 years' experience use AI frequently for polishing, versus 36.1% of 20+ year veterans, 22.5% of whom refuse AI entirely.
Sources (1)
AI4S势不可挡！Nature重磅报告：调研上万科学家，超4成自费也要用AI
TL;DR - Fudan University, the Shanghai Academy of AI for Science, and Springer Nature surveyed 10,480 researchers across 117 countries and 12 disciplines on AI use in research — the largest such study to date — finding AI is now embedded in daily scientific work but blocked from trusted, accountability-bearing tasks by a persistent credibility gap.
- Usage splits by task type: high adoption in text/information work (manuscript polishing 44.6%, literature search 43.8%, research planning 40.9%) versus low adoption where accountability matters (peer review 23.5%, submission decisions 25.7%).
- General-purpose LLMs dominate at 75.9% of tool mentions, with specialized literature/data/visualization tools under 10% combined; ChatGPT leads globally at 36.8% mention rate, while DeepSeek reaches 48.8% among Chinese researchers but under 10% abroad.
- Institutional funding is scarce: only 11.1% get AI via institutional procurement, 45.4% use free tiers, and 41.3% pay out of pocket — highest self-funding among hospital staff (49.1%) — creating compliance blind spots from unmonitored personal accounts.
- Open-ended responses show concern outweighing enthusiasm: accuracy/hallucination drew 4,935 mentions versus 3,442 for productivity gains, followed by weakened critical thinking (1,167), integrity damage from fabricated content (702), and data/IP leakage (216) — a paradox given that citation traceability is the top cited benefit.
- Generational divide: 46.8% of researchers with under 3 years' experience use AI frequently for polishing, versus 36.1% of 20+ year veterans, 22.5% of whom refuse AI entirely.