When AI Meets Science: Research Diversity, Interdisciplinarity, Visibility, and Retractions across Disciplines in a Global Surge

A comprehensive study reveals that while AI adoption in research has surged exponentially since 2015, the technology remains concentrated in narrow domains tied to computer science with limited epistemological transformation. The research identifies concerning patterns including higher retraction rates in AI-supported work, citation inflation, and geographic disparities in adoption across countries and disciplines.

This research presents a sobering assessment of AI's actual impact on scientific progress, challenging the narrative of transformative technological revolution. The study documents that post-2015 AI adoption multiplied at least fourfold across all research domains, yet this explosive growth masks fundamental limitations in how broadly and meaningfully AI is reshaping scientific inquiry. The concentration of AI-supported research within computer science-adjacent fields and traditional statistical methodologies suggests the technology functions more as an optimization tool than a paradigm shifter, leaving vast swaths of scientific investigation untouched.

The data reveals critical quality concerns that warrant serious attention. AI-supported research exhibits substantially higher retraction rates and benefits from disproportionate citation premiums unrelated to actual scientific merit. These patterns suggest potential issues with reproducibility, verification, and peer review standards in AI-driven research. The geographic analysis uncovers pronounced heterogeneity, with middle-income Asian nations, particularly China, accelerating their AI research capabilities while traditional Western scientific powers maintain dominance in certain domains.

For the research and academic communities, these findings underscore urgent needs for enhanced transparency, reproducibility standards, and ethical guidelines governing AI tool deployment. The uneven adoption landscape creates competitive pressures that may incentivize researchers to adopt AI not for scientific necessity but for citation advantage. This dynamic threatens research integrity and resource allocation efficiency. Moving forward, scientific institutions must establish stronger methodological standards and verification protocols specifically for AI-supported work while ensuring equitable access to AI tools across geographic regions and scientific disciplines to unlock genuine transformative potential.

• →AI research adoption exploded after 2015 but remains confined to computer-science-adjacent fields with limited transformational impact on epistemology
• →AI-supported studies show significantly higher retraction rates and receive unwarranted citation bonuses compared to non-AI work across most disciplines
• →Geographic disparities in AI adoption are widening, with middle-income Asian countries accelerating capabilities while traditional Western research dominance persists
• →Current AI integration in science raises serious concerns about reproducibility, transparency, and ethics requiring stronger institutional standards
• →AI's transformative potential in research remains largely untapped due to narrow deployment patterns and quality control issues