When AI Meets Science: Research Diversity, Interdisciplinarity, Visibility, and Retractions across Disciplines in a Global Surge
AI adoption surged post-2015, but papers show limited innovation and more retractions...
A new preprint from researchers Andrés F. Castro Torres, Joan Giner-Miguelez, and Mercè Crosàs (arXiv:2605.06033) examines how artificial intelligence has permeated scientific research globally from 1960 to 2015. The study documents that prior to 2015, AI adoption varied significantly by country and field, with chemistry benefiting earliest. After 2015, the landscape shifted dramatically: AI-supported works grew exponentially across all disciplines, at least quadrupling in volume. Yet the authors argue this surge has not triggered a true paradigm shift—most AI-assisted research remains confined to a few topics tightly linked to computer science and conventional statistical methods, limiting epistemological transformation.
More troubling, the paper reveals two major red flags. First, AI-supported papers enjoy an unwarranted citation premium, meaning they are cited more than their quality or novelty justifies. Second, retraction rates for AI-assisted work are substantially higher than for non-AI research across most fields, hinting at issues in reproducibility, transparency, or ethical oversight. Geographically, the surge is driven heavily by middle-income Asian countries, especially China. The authors emphasize that without better research practices—like open data, rigorous peer review, and ethical guidelines—AI's transformative potential in science remains untapped while its rapid adoption introduces new risks to scientific integrity.
- AI-supported research papers grew 4x across all domains after 2015, but remain narrow in topic and tied to computer science.
- AI-assisted papers have an unwarranted citation premium and show substantially higher retraction rates than non-AI works.
- Adoption concentrated in middle-income Asian countries, especially China, raising questions about global research equity.
Why It Matters
AI's rapid adoption in science risks undermining reproducibility and integrity unless transparency and ethical standards catch up.