Scientists are rapidly adopting AI, with natural-science papers mentioning the technology surging nearly 30-fold since 2010 to more than 80,000 in 2025—up 26% from 2024—according to Stanford’s Artificial Intelligence Index Report 2026. Even so, the report finds today’s autonomous AI “agents” falter on complex, multistep workflows, scoring roughly half as well as human PhD specialists. Physical sciences led in volume (33,000 papers) while Earth sciences showed the highest share of AI-mentioning work (9%), amid expanding “science foundation models” such as AION-1 for astronomy. Researchers cite widespread reliance on AI, but clear productivity gains remain unproven; some scholars, including Princeton’s Arvind Narayanan, warn that quality may be slipping as adoption outpaces evolving norms. The report also flags milestones such as AI-generated content surpassing human-authored material online in late 2024 and the debut of operational AI-powered weather forecasts in early 2025.
Related articles:
ChatGPT listed as author on research papers: many scientists disapprove
Tools such as ChatGPT threaten transparent science; here are our ground rules
Highly accurate protein structure prediction with AlphaFold
Reflexion: Language Agents with Verbal Reinforcement Learning
GPT-4 Technical Report
